Upload difficulty scorer model

__init.py__

@@ -0,0 +1 @@
+from .model import CausalLMForRegression

config.json

@@ -0,0 +1,31 @@
+{
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 12288,
+  "max_position_embeddings": 40960,
+  "max_window_layers": 36,
+  "model_type": "qwen3",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 36,
+  "num_key_value_heads": 8,
+  "output_hidden_states": true,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "temperature": 0.6,
+  "top_k": 20,
+  "top_p": 0.95,
+  "transformers_version": "4.51.3"
+}

latest

@@ -0,0 +1 @@
+global_step2538

model.py

@@ -0,0 +1,108 @@
+import os
+from typing import Optional
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+import torch.nn as nn
+
+
+# Define a custom model that wraps a causal LM and adds a regression head
+class CausalLMForRegression(nn.Module):
+    def __init__(self, model_name):
+        super().__init__()
+        # Load the causal LM with hidden states enabled
+        self.model = AutoModelForCausalLM.from_pretrained(
+            model_name, 
+            output_hidden_states=True
+        )
+        self.base_model = "Qwen/Qwen3-8B" # WARNING: USED FOR GETTING TOKENIZER. THIS IS HARDCODED FOR NOW!!
+        # Using pooled hidden state to a single scalar
+        self.regression_head = nn.Linear(self.model.config.hidden_size, 1)
+        try:
+            regression_head_path = os.path.join(model_name, "regression_head.bin")
+            state = torch.load(regression_head_path, map_location="cpu")
+            self.regression_head.load_state_dict(state)
+        except FileNotFoundError:
+            print(f"No regression head found. Initializing with random weights!")
+        self._keys_to_ignore_on_save = []
+
+    def forward(self, input_ids, attention_mask=None, labels=None):
+        # Flatten extra dimensions if present
+        if input_ids.dim() == 3:
+            # e.g. from (accum_steps, batch_size, seq_length) to (accum_steps * batch_size, seq_length)
+            input_ids = input_ids.view(-1, input_ids.size(-1))
+        if attention_mask is not None and attention_mask.dim() == 3:
+            attention_mask = attention_mask.view(-1, attention_mask.size(-1))
+
+        outputs = self.model(input_ids, attention_mask=attention_mask)
+        hidden_states = outputs.hidden_states[-1]  # Now should have shape: (batch, seq_length, hidden_size)
+        
+        # Mean-pooling over non-padding tokens
+        if attention_mask is not None:
+            mask = attention_mask.unsqueeze(-1).expand_as(hidden_states).to(hidden_states.dtype)
+            hidden_sum = torch.sum(hidden_states * mask, dim=1)
+            lengths = mask.sum(dim=1)
+            pooled = hidden_sum / lengths
+        else:
+            pooled = hidden_states.mean(dim=1)
+        
+        logits = self.regression_head(pooled).squeeze(-1)
+        
+        loss = None
+        if labels is not None:
+            loss_fn = nn.HuberLoss() #nn.MSELoss()
+            loss = loss_fn(logits, labels)
+        
+        return {"loss": loss, "logits": logits}
+
+    def get_input_embeddings(self):
+        # Delegate to the underlying causal LM's get_input_embeddings method.
+        return self.model.get_input_embeddings()
+
+    def save_pretrained(self, output_dir, safe_serialization=False):
+        os.makedirs(output_dir, exist_ok=True)
+
+        # Ensure we are saving the entire model properly
+        model_state_dict = self.model.state_dict()
+        for key, value in model_state_dict.items():
+            if value.shape[0] == 0:
+                print(f"Warning: Tensor {key} has shape {value.shape}, which may be problematic.")
+
+        # Save model with proper weight tie handling
+        self.model.save_pretrained(output_dir, safe_serialization=False)
+        torch.save(self.regression_head.state_dict(), os.path.join(output_dir, "regression_head.bin"))
+
+
+    def get_tokenizer(self):
+        try:
+            tokenizer = AutoTokenizer.from_pretrained(self.model.name_or_path)
+            print(f"Loaded tokenizer from {self.model.name_or_path}")
+        except:
+            tokenizer = AutoTokenizer.from_pretrained(self.base_model)
+            print(f"Loaded tokenizer from {self.base_model}")
+        return tokenizer
+
+    @classmethod
+    def from_pretrained(cls, output_dir):
+        from_local = os.path.exists(output_dir)
+        loading_kwargs = {"use_safetensors": False} if from_local else {}
+        
+        model = AutoModelForCausalLM.from_pretrained(output_dir, **loading_kwargs)
+
+        # Explicitly enable `output_hidden_states` after loading
+        model.config.output_hidden_states = True  
+
+        # Create an uninitialized instance of CausalLMForRegression
+        instance = cls.__new__(cls)
+        nn.Module.__init__(instance)
+        instance._keys_to_ignore_on_save = []
+        instance.model = model
+
+        # Load the regression head separately
+        instance.regression_head = nn.Linear(model.config.hidden_size, 1)
+        try:
+            regression_head_path = os.path.join(output_dir, "regression_head.bin")
+            state = torch.load(regression_head_path, map_location="cpu")
+            instance.regression_head.load_state_dict(state)
+        except FileNotFoundError:
+            print(f"No regression head found. Initializing with random weights!")
+        return instance

pytorch_model-00001-of-00002.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e97747b78d7635f05b471e0a518f8f8b5add4fbdca1e38b14ca69b81795f5db5
+size 16380934585

pytorch_model-00002-of-00002.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0dbe39d5f09210d37da3795026145a1666b6b401aede8ac7a3cf0da3de9bfefd
+size 600005

pytorch_model.bin.index.json

@@ -0,0 +1,406 @@
+{
+  "metadata": {
+    "total_size": 16381478928
+  },
+  "weight_map": {
+    "lm_head.weight": "pytorch_model-00001-of-00002.bin",
+    "model.embed_tokens.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.0.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.0.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.0.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.0.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.0.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.0.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.1.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.1.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.1.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.1.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.1.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.1.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.1.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.10.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.10.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.10.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.10.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.10.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.10.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.10.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.10.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.10.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.10.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.10.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.11.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.11.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.11.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.11.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.11.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.11.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.11.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.11.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.11.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.11.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.11.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.12.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.12.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.12.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.12.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.12.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.12.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.12.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.12.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.12.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.12.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.12.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.13.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.13.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.13.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.13.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.13.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.13.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.13.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.13.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.13.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.13.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.13.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.14.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.14.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.14.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.14.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.14.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.14.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.14.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.14.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.14.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.14.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.14.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.15.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.15.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.15.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.15.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.15.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.15.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.15.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.15.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.15.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.15.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.15.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.16.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.16.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.16.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.16.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.16.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.16.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.16.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.16.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.16.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.16.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.16.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.17.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.17.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.17.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.17.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.17.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.17.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.17.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.17.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.17.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.17.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.17.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.18.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.18.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.18.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.18.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.18.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.18.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.18.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.18.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.18.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.18.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.18.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.19.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.19.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.19.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.19.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.19.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.19.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.19.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.19.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.19.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.19.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.19.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.2.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.2.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.2.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.2.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.2.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.2.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.2.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.2.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.2.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.2.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.2.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.20.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.20.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.20.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.20.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.20.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.20.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.20.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.20.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.20.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.20.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.20.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.21.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.21.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.21.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.21.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.21.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.21.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.21.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.21.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.21.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.21.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.21.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.22.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.22.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.22.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.22.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.22.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.22.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.22.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.22.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.22.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.22.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.22.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.23.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.23.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.23.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.23.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.23.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.23.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.23.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.23.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.23.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.23.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.23.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.24.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.24.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.24.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.24.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.24.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.24.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.24.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.24.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.24.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.24.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.24.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.25.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.25.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.25.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.25.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.25.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.25.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.25.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.25.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.25.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.25.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.25.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.26.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.26.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.26.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.26.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.26.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.26.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.26.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.26.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.26.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.26.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.26.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.27.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.27.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.27.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.27.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.27.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.27.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.27.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.27.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.27.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.27.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.27.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.28.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.28.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.28.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.28.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.28.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.28.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.28.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.28.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.28.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.28.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.28.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.29.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.29.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.29.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.29.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.29.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.29.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.29.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.29.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.29.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.29.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.29.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.3.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.3.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.3.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.3.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.3.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.3.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.3.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.3.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.3.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.3.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.3.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.30.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.30.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.30.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.30.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.30.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.30.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.30.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.30.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.30.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.30.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.30.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.31.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.31.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.31.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.31.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.31.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.31.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.31.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.31.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.31.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.31.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.31.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.32.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.32.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.32.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.32.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.32.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.32.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.32.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.32.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.32.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.32.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.32.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.33.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.33.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.33.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.33.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.33.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.33.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.33.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.33.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.33.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.33.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.33.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.34.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.34.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.34.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.34.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.34.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.34.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.34.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.34.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.34.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.34.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.34.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.35.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.35.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.35.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.35.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.35.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.35.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.35.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.35.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.35.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.35.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.35.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.4.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.4.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.4.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.4.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.4.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.4.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.4.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.4.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.4.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.4.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.4.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.5.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.5.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.5.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.5.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.5.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.5.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.5.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.5.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.5.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.5.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.5.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.6.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.6.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.6.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.6.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.6.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.6.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.6.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.6.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.6.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.6.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.6.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.7.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.7.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.7.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.7.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.7.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.7.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.7.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.7.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.7.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.7.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.7.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.8.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.8.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.8.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.8.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.8.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.8.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.8.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.8.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.8.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.8.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.8.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.9.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.9.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.9.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.9.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.9.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
+    "model.layers.9.self_attn.k_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.9.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.9.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.9.self_attn.q_norm.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.9.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.layers.9.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
+    "model.norm.weight": "pytorch_model-00001-of-00002.bin"
+  }
+}

regression_head.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bfab33e6e3e4560d39f766bcde7843cbbc3a0f61175e99922eed5dae050e39da
+size 16381480584

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2658016c0b7dd2e81bdc3fd1a7c273d1a3e71d5bbc85e15564086266c450fb06
+size 14960

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d06145b81b5ded05d055df7243b4b2581bb23f0bb853a0534d6e4a62cb6d210
+size 14960

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed43f565759da9b9bb7102cd17708bd0bcc140f30d1d2b843fb63571e7eed2a8
+size 14960

rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f418ab8bae8f48f01ab52c6d5fc505596f51a69b886850493bbc0092968d129d
+size 14960

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c510c43d10855e06cc544329b7ff4b0d2b08a8116db99cde31586911f6d4a39c
+size 1064

trainer_state.json

@@ -0,0 +1,484 @@
+{
+  "best_global_step": 1500,
+  "best_metric": 0.6232942938804626,
+  "best_model_checkpoint": "/data/horse/ws/luwe911g-nemotron_ws/synthetic-instruction-data-generation/analysis/train_difficulty_scorer/checkpoints/checkpoint-1500",
+  "epoch": 7.9774111176586135,
+  "eval_steps": 100,
+  "global_step": 2544,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.3142801021410332,
+      "grad_norm": 2.72916316986084,
+      "learning_rate": 9.9e-06,
+      "loss": 0.062,
+      "step": 100
+    },
+    {
+      "epoch": 0.3142801021410332,
+      "eval_loss": 0.01668722555041313,
+      "eval_mse": 0.03342443332076073,
+      "eval_pearson_r": 0.2520392835140228,
+      "eval_r2": -0.12196338176727295,
+      "eval_runtime": 0.9873,
+      "eval_samples_per_second": 104.321,
+      "eval_steps_per_second": 26.334,
+      "step": 100
+    },
+    {
+      "epoch": 0.6285602042820664,
+      "grad_norm": 7.041975021362305,
+      "learning_rate": 9.594926350245501e-06,
+      "loss": 0.0183,
+      "step": 200
+    },
+    {
+      "epoch": 0.6285602042820664,
+      "eval_loss": 0.013392927125096321,
+      "eval_mse": 0.026777038350701332,
+      "eval_pearson_r": 0.35883840918540955,
+      "eval_r2": 0.10117077827453613,
+      "eval_runtime": 0.9862,
+      "eval_samples_per_second": 104.441,
+      "eval_steps_per_second": 26.364,
+      "step": 200
+    },
+    {
+      "epoch": 0.9428403064230996,
+      "grad_norm": 6.196834087371826,
+      "learning_rate": 9.185761047463177e-06,
+      "loss": 0.017,
+      "step": 300
+    },
+    {
+      "epoch": 0.9428403064230996,
+      "eval_loss": 0.014142508618533611,
+      "eval_mse": 0.028310200199484825,
+      "eval_pearson_r": 0.34453094005584717,
+      "eval_r2": 0.04970693588256836,
+      "eval_runtime": 0.9869,
+      "eval_samples_per_second": 104.363,
+      "eval_steps_per_second": 26.344,
+      "step": 300
+    },
+    {
+      "epoch": 1.2545668827342369,
+      "grad_norm": 3.1854069232940674,
+      "learning_rate": 8.776595744680852e-06,
+      "loss": 0.013,
+      "step": 400
+    },
+    {
+      "epoch": 1.2545668827342369,
+      "eval_loss": 0.013145764358341694,
+      "eval_mse": 0.02628500573337078,
+      "eval_pearson_r": 0.3917391896247864,
+      "eval_r2": 0.11768698692321777,
+      "eval_runtime": 0.9902,
+      "eval_samples_per_second": 104.021,
+      "eval_steps_per_second": 26.258,
+      "step": 400
+    },
+    {
+      "epoch": 1.56884698487527,
+      "grad_norm": 2.2994000911712646,
+      "learning_rate": 8.367430441898528e-06,
+      "loss": 0.0135,
+      "step": 500
+    },
+    {
+      "epoch": 1.56884698487527,
+      "eval_loss": 0.011843894608318806,
+      "eval_mse": 0.023691682144999504,
+      "eval_pearson_r": 0.4777704179286957,
+      "eval_r2": 0.20473748445510864,
+      "eval_runtime": 0.9867,
+      "eval_samples_per_second": 104.389,
+      "eval_steps_per_second": 26.351,
+      "step": 500
+    },
+    {
+      "epoch": 1.8831270870163033,
+      "grad_norm": 1.4870805740356445,
+      "learning_rate": 7.958265139116204e-06,
+      "loss": 0.0115,
+      "step": 600
+    },
+    {
+      "epoch": 1.8831270870163033,
+      "eval_loss": 0.013224405236542225,
+      "eval_mse": 0.026459576562047005,
+      "eval_pearson_r": 0.4158453345298767,
+      "eval_r2": 0.11182713508605957,
+      "eval_runtime": 1.0077,
+      "eval_samples_per_second": 102.216,
+      "eval_steps_per_second": 25.802,
+      "step": 600
+    },
+    {
+      "epoch": 2.1948536633274407,
+      "grad_norm": 4.134687900543213,
+      "learning_rate": 7.549099836333879e-06,
+      "loss": 0.0088,
+      "step": 700
+    },
+    {
+      "epoch": 2.1948536633274407,
+      "eval_loss": 0.013537898659706116,
+      "eval_mse": 0.027076171711087227,
+      "eval_pearson_r": 0.4764866530895233,
+      "eval_r2": 0.09112972021102905,
+      "eval_runtime": 0.9823,
+      "eval_samples_per_second": 104.858,
+      "eval_steps_per_second": 26.469,
+      "step": 700
+    },
+    {
+      "epoch": 2.5091337654684738,
+      "grad_norm": 0.8281658291816711,
+      "learning_rate": 7.1399345335515555e-06,
+      "loss": 0.0064,
+      "step": 800
+    },
+    {
+      "epoch": 2.5091337654684738,
+      "eval_loss": 0.010303011164069176,
+      "eval_mse": 0.020632537081837654,
+      "eval_pearson_r": 0.5958652496337891,
+      "eval_r2": 0.30742424726486206,
+      "eval_runtime": 0.9884,
+      "eval_samples_per_second": 104.209,
+      "eval_steps_per_second": 26.305,
+      "step": 800
+    },
+    {
+      "epoch": 2.823413867609507,
+      "grad_norm": 4.725385665893555,
+      "learning_rate": 6.730769230769232e-06,
+      "loss": 0.0068,
+      "step": 900
+    },
+    {
+      "epoch": 2.823413867609507,
+      "eval_loss": 0.012674727477133274,
+      "eval_mse": 0.025358131155371666,
+      "eval_pearson_r": 0.5820975303649902,
+      "eval_r2": 0.14879947900772095,
+      "eval_runtime": 0.9974,
+      "eval_samples_per_second": 103.266,
+      "eval_steps_per_second": 26.067,
+      "step": 900
+    },
+    {
+      "epoch": 3.135140443920644,
+      "grad_norm": 0.9976857900619507,
+      "learning_rate": 6.321603927986907e-06,
+      "loss": 0.0051,
+      "step": 1000
+    },
+    {
+      "epoch": 3.135140443920644,
+      "eval_loss": 0.009650396183133125,
+      "eval_mse": 0.019312988966703415,
+      "eval_pearson_r": 0.6038205623626709,
+      "eval_r2": 0.3517177700996399,
+      "eval_runtime": 0.9924,
+      "eval_samples_per_second": 103.788,
+      "eval_steps_per_second": 26.199,
+      "step": 1000
+    },
+    {
+      "epoch": 3.4494205460616776,
+      "grad_norm": 1.735731840133667,
+      "learning_rate": 5.912438625204583e-06,
+      "loss": 0.0034,
+      "step": 1100
+    },
+    {
+      "epoch": 3.4494205460616776,
+      "eval_loss": 0.011387597769498825,
+      "eval_mse": 0.022768154740333557,
+      "eval_pearson_r": 0.5331981778144836,
+      "eval_r2": 0.2357376217842102,
+      "eval_runtime": 1.0408,
+      "eval_samples_per_second": 98.965,
+      "eval_steps_per_second": 24.981,
+      "step": 1100
+    },
+    {
+      "epoch": 3.7637006482027107,
+      "grad_norm": 0.24521802365779877,
+      "learning_rate": 5.503273322422259e-06,
+      "loss": 0.0032,
+      "step": 1200
+    },
+    {
+      "epoch": 3.7637006482027107,
+      "eval_loss": 0.010628500953316689,
+      "eval_mse": 0.02123476378619671,
+      "eval_pearson_r": 0.5781332850456238,
+      "eval_r2": 0.2872092127799988,
+      "eval_runtime": 0.9896,
+      "eval_samples_per_second": 104.08,
+      "eval_steps_per_second": 26.273,
+      "step": 1200
+    },
+    {
+      "epoch": 4.075427224513848,
+      "grad_norm": 1.210878610610962,
+      "learning_rate": 5.094108019639935e-06,
+      "loss": 0.0025,
+      "step": 1300
+    },
+    {
+      "epoch": 4.075427224513848,
+      "eval_loss": 0.010471797548234463,
+      "eval_mse": 0.020952731370925903,
+      "eval_pearson_r": 0.5631570219993591,
+      "eval_r2": 0.2966762185096741,
+      "eval_runtime": 0.9887,
+      "eval_samples_per_second": 104.175,
+      "eval_steps_per_second": 26.297,
+      "step": 1300
+    },
+    {
+      "epoch": 4.389707326654881,
+      "grad_norm": 0.7935001254081726,
+      "learning_rate": 4.684942716857611e-06,
+      "loss": 0.0016,
+      "step": 1400
+    },
+    {
+      "epoch": 4.389707326654881,
+      "eval_loss": 0.01045281533151865,
+      "eval_mse": 0.020906325429677963,
+      "eval_pearson_r": 0.5742180347442627,
+      "eval_r2": 0.2982339859008789,
+      "eval_runtime": 0.9895,
+      "eval_samples_per_second": 104.088,
+      "eval_steps_per_second": 26.275,
+      "step": 1400
+    },
+    {
+      "epoch": 4.703987428795914,
+      "grad_norm": 0.7895488739013672,
+      "learning_rate": 4.275777414075287e-06,
+      "loss": 0.0015,
+      "step": 1500
+    },
+    {
+      "epoch": 4.703987428795914,
+      "eval_loss": 0.009268084540963173,
+      "eval_mse": 0.01854919083416462,
+      "eval_pearson_r": 0.6232942938804626,
+      "eval_r2": 0.3773563504219055,
+      "eval_runtime": 0.9876,
+      "eval_samples_per_second": 104.292,
+      "eval_steps_per_second": 26.326,
+      "step": 1500
+    },
+    {
+      "epoch": 5.015714005107052,
+      "grad_norm": 0.3999693691730499,
+      "learning_rate": 3.866612111292962e-06,
+      "loss": 0.0014,
+      "step": 1600
+    },
+    {
+      "epoch": 5.015714005107052,
+      "eval_loss": 0.009675208479166031,
+      "eval_mse": 0.01935533992946148,
+      "eval_pearson_r": 0.5987623333930969,
+      "eval_r2": 0.35029613971710205,
+      "eval_runtime": 1.0789,
+      "eval_samples_per_second": 95.47,
+      "eval_steps_per_second": 24.099,
+      "step": 1600
+    },
+    {
+      "epoch": 5.329994107248085,
+      "grad_norm": 0.2833040952682495,
+      "learning_rate": 3.457446808510639e-06,
+      "loss": 0.0006,
+      "step": 1700
+    },
+    {
+      "epoch": 5.329994107248085,
+      "eval_loss": 0.00994726363569498,
+      "eval_mse": 0.019889621064066887,
+      "eval_pearson_r": 0.5999482870101929,
+      "eval_r2": 0.3323618769645691,
+      "eval_runtime": 0.9912,
+      "eval_samples_per_second": 103.917,
+      "eval_steps_per_second": 26.231,
+      "step": 1700
+    },
+    {
+      "epoch": 5.644274209389118,
+      "grad_norm": 0.1224384605884552,
+      "learning_rate": 3.0482815057283143e-06,
+      "loss": 0.0006,
+      "step": 1800
+    },
+    {
+      "epoch": 5.644274209389118,
+      "eval_loss": 0.009732786566019058,
+      "eval_mse": 0.01945064589381218,
+      "eval_pearson_r": 0.6066421866416931,
+      "eval_r2": 0.3470969796180725,
+      "eval_runtime": 0.9881,
+      "eval_samples_per_second": 104.236,
+      "eval_steps_per_second": 26.312,
+      "step": 1800
+    },
+    {
+      "epoch": 5.958554311530151,
+      "grad_norm": 0.03875954821705818,
+      "learning_rate": 2.6391162029459905e-06,
+      "loss": 0.0007,
+      "step": 1900
+    },
+    {
+      "epoch": 5.958554311530151,
+      "eval_loss": 0.009639283642172813,
+      "eval_mse": 0.019304735586047173,
+      "eval_pearson_r": 0.6057748198509216,
+      "eval_r2": 0.35199475288391113,
+      "eval_runtime": 1.0009,
+      "eval_samples_per_second": 102.908,
+      "eval_steps_per_second": 25.977,
+      "step": 1900
+    },
+    {
+      "epoch": 6.270280887841288,
+      "grad_norm": 0.4777383804321289,
+      "learning_rate": 2.2299509001636663e-06,
+      "loss": 0.0003,
+      "step": 2000
+    },
+    {
+      "epoch": 6.270280887841288,
+      "eval_loss": 0.00959899090230465,
+      "eval_mse": 0.01921091228723526,
+      "eval_pearson_r": 0.6080344915390015,
+      "eval_r2": 0.355144202709198,
+      "eval_runtime": 0.9877,
+      "eval_samples_per_second": 104.285,
+      "eval_steps_per_second": 26.324,
+      "step": 2000
+    },
+    {
+      "epoch": 6.584560989982322,
+      "grad_norm": 0.1112111359834671,
+      "learning_rate": 1.820785597381342e-06,
+      "loss": 0.0002,
+      "step": 2100
+    },
+    {
+      "epoch": 6.584560989982322,
+      "eval_loss": 0.009637333452701569,
+      "eval_mse": 0.01925988681614399,
+      "eval_pearson_r": 0.6114506721496582,
+      "eval_r2": 0.35350024700164795,
+      "eval_runtime": 1.0749,
+      "eval_samples_per_second": 95.819,
+      "eval_steps_per_second": 24.187,
+      "step": 2100
+    },
+    {
+      "epoch": 6.898841092123355,
+      "grad_norm": 0.32920289039611816,
+      "learning_rate": 1.4116202945990182e-06,
+      "loss": 0.0002,
+      "step": 2200
+    },
+    {
+      "epoch": 6.898841092123355,
+      "eval_loss": 0.009636594913899899,
+      "eval_mse": 0.019291426986455917,
+      "eval_pearson_r": 0.6066728830337524,
+      "eval_r2": 0.35244154930114746,
+      "eval_runtime": 0.9867,
+      "eval_samples_per_second": 104.391,
+      "eval_steps_per_second": 26.351,
+      "step": 2200
+    },
+    {
+      "epoch": 7.210567668434492,
+      "grad_norm": 0.1292334944009781,
+      "learning_rate": 1.002454991816694e-06,
+      "loss": 0.0001,
+      "step": 2300
+    },
+    {
+      "epoch": 7.210567668434492,
+      "eval_loss": 0.009695847518742085,
+      "eval_mse": 0.01940760761499405,
+      "eval_pearson_r": 0.6034567952156067,
+      "eval_r2": 0.3485416769981384,
+      "eval_runtime": 1.0359,
+      "eval_samples_per_second": 99.427,
+      "eval_steps_per_second": 25.098,
+      "step": 2300
+    },
+    {
+      "epoch": 7.5248477705755255,
+      "grad_norm": 0.01900539919734001,
+      "learning_rate": 5.932896890343699e-07,
+      "loss": 0.0001,
+      "step": 2400
+    },
+    {
+      "epoch": 7.5248477705755255,
+      "eval_loss": 0.0095865149050951,
+      "eval_mse": 0.019158780574798584,
+      "eval_pearson_r": 0.6077873706817627,
+      "eval_r2": 0.3568941354751587,
+      "eval_runtime": 0.9886,
+      "eval_samples_per_second": 104.191,
+      "eval_steps_per_second": 26.301,
+      "step": 2400
+    },
+    {
+      "epoch": 7.839127872716559,
+      "grad_norm": 0.021035829558968544,
+      "learning_rate": 1.8412438625204584e-07,
+      "loss": 0.0001,
+      "step": 2500
+    },
+    {
+      "epoch": 7.839127872716559,
+      "eval_loss": 0.009597421623766422,
+      "eval_mse": 0.019200436770915985,
+      "eval_pearson_r": 0.6084479093551636,
+      "eval_r2": 0.355495810508728,
+      "eval_runtime": 0.987,
+      "eval_samples_per_second": 104.357,
+      "eval_steps_per_second": 26.342,
+      "step": 2500
+    }
+  ],
+  "logging_steps": 100,
+  "max_steps": 2544,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 8,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+#   python zero_to_fp32.py . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device, weights_only=False)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+    total_files = len(files)
+    state_dicts = []
+    for f in tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                                          tag,
+                                                          exclude_frozen_parameters,
+                                                          lazy_mode=True)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument("output_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+                                               args.output_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)