Upload runs/LFM2.5-VL-450M-vlm_sft-vlm_mix_pa-all-lr1em05-w0p1-lora_a-20260502_104403/README.md with huggingface_hub
Browse files
runs/LFM2.5-VL-450M-vlm_sft-vlm_mix_pa-all-lr1em05-w0p1-lora_a-20260502_104403/README.md
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Training artifacts
|
| 2 |
+
|
| 3 |
+
Local run directory: `/data/nutonic/sat_vl_out/LFM2.5-VL-450M-vlm_sft-vlm_mix_pa-all-lr1em05-w0p1-lora_a-20260502_104403`
|
| 4 |
+
|
| 5 |
+
Hub layout under `runs/LFM2.5-VL-450M-vlm_sft-vlm_mix_pa-all-lr1em05-w0p1-lora_a-20260502_104403/`:
|
| 6 |
+
|
| 7 |
+
- `checkpoints/<checkpoint_dir_name>/` — Hugging Face Trainer / LEAP checkpoints (LoRA + optimizer state).
|
| 8 |
+
- `merged_after_training/` — full-precision merged model snapshot after the single training run.
|
| 9 |
+
- `configs/` — generated LEAP YAML for this run.
|
| 10 |
+
- `manifest.jsonl` — one JSON line with epoch/LR/dataset summary.
|
| 11 |
+
- `ray_logs/` — optional Ray Train experiment export if enabled with ``--upload-ray-logs``.
|
| 12 |
+
|
| 13 |
+
The **canonical release** weights may also be uploaded separately to your primary model repo (repo root).
|