re adds the join us shield

updates gradio versions

updates readmes

Update README.md

adds how to instructions

adds image and title to the interface

adds tabbed interface, advanced mode, connectors

adds interface linkage so updates work correctly

adds interface.py improvements to the flow and tab

improves interface flow

improves interface.py

adds a lot of improvements to launch.sh and demo spaces and secures read token at demo launch

adds hf compatible readme yaml

adds hf compatible readme yaml

improves spaces deployment , configuration for custom settings , adds interface for spaces deployment

adds steps logging + parsing - woops

adds experiment id fix

adds normalized variables on update

propagates changes

adds improved dataset utils with new interface

adds packing argument to gpt oss custom config

fixes circular import

add configs to _init_.py

solves import error

non destructive dataset operations

adds imports and forces monitoring

adds scheduler stuff and hopes for the best with track tonic

adds local and remote training monitors to config

improve dataset utils non destructive writes

adds parameters to medical config

adds parameters to medical config

adds parameters to medical config

adds improved dev and system prompts to o1-medical

adds improved launch for reasoning gpt-oss configs and new config for medical reasoning

adds improved dataset utils in tracktonic

adds optimizations for faster training

adds a100 memory optimized

adds template files , adds non destructive dataset updates

adds flash attention 3 kernel

adapt lr_scheduler according to trl version

sets min_lr

increases batchsize and gradient accumulation steps in memory optimized

increases max samples in memory optimized

adds better launch.sh and eval / test splits auto

adds repoid only based on repo name, adds version-robust sfttrainer

coerce akk numeric config values to safe values

adds defensive programming (boo) and adaptations based on transformer versions

improves launcher with model family and defaults based on options, updates trl trainer , removes trl config paths by switching to trainingarguments class , tokenizer parameters updated to sfttrainer , resolves evaluation_strategy error

hide all tokens in logs, never persist to disk, remove max_seq_length from config, add to trainer

changes default model repo name + adds non default option