cmedina

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-03 01:24:27 -04:00

6f45390be8 fix: remove quantization_config from config.json to prevent re-quantization

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-03 01:22:55 -04:00

1c3da52028 feat: create proper index file from safetensors metadata

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-03 01:19:56 -04:00

b15bc5c975 feat: detect pre-quantized model and skip re-quantization

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-03 00:26:40 -04:00

4308fd3391 feat: continuous processing - add to GPU with fewer tasks (max 4 per GPU)

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-03 00:23:23 -04:00

8a30ec92c6 fix: reduce to 3 shards per GPU (6 total) for memory

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-03 00:16:17 -04:00

23dcdd2c6f feat: rename 0-indexed saved files to 1-indexed

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-03 00:15:40 -04:00

b4538c013b fix: use 1-indexing for shard names (model-00001-of-00016)

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-03 00:14:59 -04:00

50623cdcd0 fix: use 0-indexing consistently (shard 0-15)

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 23:39:25 -04:00

262b1181b7 fix: skip to first unsaved shard, fix GPU assignment

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 23:32:40 -04:00

b742b5549e fix: alternate GPUs for each shard (0,1,0,1)

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 23:27:16 -04:00

78817532ad fix: import concurrent.futures at top level

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 23:24:52 -04:00

bc3bc6d6eb feat: process 8 shards at once (4 per GPU) with better logging

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 23:12:06 -04:00

9e2c7bec8d fix: remove duplicate sequential code

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 23:05:59 -04:00

7040c84618 feat: process 2 shards at a time (one per GPU)

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 23:04:53 -04:00

d698a32f0f feat: skip already processed shards (resume support)

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 22:52:51 -04:00

d1f0c81ac4 fix: remove quant_states reference

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 22:48:11 -04:00

18ef0f73ef fix: don't save quant_state objects (BnB rebuilds on load)

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 22:43:29 -04:00

8c13be4d1f fix: free GPU memory after each layer quantization

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 22:41:54 -04:00

b9f6ddaf50 fix: free GPU memory after each quantization

cmedina pushed to main at cmedina/agenx-lora-training

2026-07-02 22:40:33 -04:00

1751d33f69 fix: load shard to CPU, quantize on GPU