cmedina

0 Followers · 0 Following

• Joined on 2026-05-22

Repositories

1

Projects Packages Public Activity Starred Repositories

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 21:05:49 -04:00

ed0b1eddf8 fix: use messages format for SFTTrainer (system/user/assistant)

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:59:25 -04:00

d8ad31fd03 fix: remove max_seq_length from SFTTrainer (newer TRL API)

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:48:08 -04:00

8ad87f0e7e fix: remove dataset_text_field (dataset pre-formatted)

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:41:55 -04:00

6bd6662f5f docs: rename config and docs to reflect Ornith-35B model

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:39:30 -04:00

6f31fa0197 feat: add --check-only flag to validate setup without training

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:38:32 -04:00

83ba987840 fix: tokenizer -> processing_class SFTTrainer API change

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:34:07 -04:00

9d5d4eebc7 fix: mixed_precision -> bf16/api change

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:29:04 -04:00

9280103260 fix: evaluation_strategy -> eval_strategy API change

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:24:45 -04:00

50fa171f68 fix: use local model path instead of HuggingFace download

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:17:05 -04:00

7f7467ef89 fix: remove max_seq_length from TrainingArguments (passed to SFTTrainer)

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 20:01:06 -04:00

7181acc580 fix: save train/test.jsonl to output_dir not parent

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 19:54:51 -04:00

d5ba471b98 fix: auto-split dataset if split files missing

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 19:01:40 -04:00

bbdbe91871 fix: fix dataset path calculation

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 18:49:41 -04:00

14a302dc5e fix: make scripts executable

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 18:48:48 -04:00

97004261f5 fix: multiple bug fixes - remove unused imports, fix typo, fix dataset paths, add ai command fallback

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 18:45:16 -04:00

75aeeeb20a fix: use torchrun for distributed training, load model on CPU

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 18:30:25 -04:00

5ad5518ca6 fix: add deepspeed to auto-install dependencies

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 18:26:50 -04:00

4f2f8ef03f feat: use DeepSpeed ZeRO-3 for optimal 2-GPU training

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 18:26:09 -04:00

a04bf4a9cf feat: add run_with_accelerate.sh helper script

cmedina pushed to main at cmedina/agenx-lora-training 2026-06-30 18:23:35 -04:00

cd58ffd31e fix: use FSDP for 2-GPU training, reduce batch size