Upload folder using huggingface_hub
Browse files- config.json +19 -0
- llm/config.json +60 -0
- llm/generation_config.json +14 -0
- llm/model.safetensors +3 -0
- lora_encoder.safetensors +3 -0
- training_metadata.json +41 -0
config.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"freeze_image_encoder": true,
|
| 3 |
+
"image_model_name": "google/siglip2-so400m-patch16-256",
|
| 4 |
+
"llm_model_name": "Qwen/Qwen3-0.6B",
|
| 5 |
+
"lora_attn_implementation": "sdpa",
|
| 6 |
+
"lora_dropout": 0.1,
|
| 7 |
+
"lora_hidden_size": 768,
|
| 8 |
+
"lora_intermediate_size": 3072,
|
| 9 |
+
"lora_mlp_ratio": 4,
|
| 10 |
+
"lora_num_attention_heads": 16,
|
| 11 |
+
"lora_num_layers": 16,
|
| 12 |
+
"lora_rank": 16,
|
| 13 |
+
"lora_target": "qkvm",
|
| 14 |
+
"lora_use_prefix": true,
|
| 15 |
+
"model_type": "vision_lora",
|
| 16 |
+
"prompt_aware": false,
|
| 17 |
+
"text_model_name": null,
|
| 18 |
+
"transformers_version": "4.57.3"
|
| 19 |
+
}
|
llm/config.json
ADDED
|
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"LoraQwen3ForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": false,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"bos_token_id": 151643,
|
| 8 |
+
"dtype": "float32",
|
| 9 |
+
"eos_token_id": 151645,
|
| 10 |
+
"head_dim": 128,
|
| 11 |
+
"hidden_act": "silu",
|
| 12 |
+
"hidden_size": 1024,
|
| 13 |
+
"initializer_range": 0.02,
|
| 14 |
+
"intermediate_size": 3072,
|
| 15 |
+
"layer_types": [
|
| 16 |
+
"full_attention",
|
| 17 |
+
"full_attention",
|
| 18 |
+
"full_attention",
|
| 19 |
+
"full_attention",
|
| 20 |
+
"full_attention",
|
| 21 |
+
"full_attention",
|
| 22 |
+
"full_attention",
|
| 23 |
+
"full_attention",
|
| 24 |
+
"full_attention",
|
| 25 |
+
"full_attention",
|
| 26 |
+
"full_attention",
|
| 27 |
+
"full_attention",
|
| 28 |
+
"full_attention",
|
| 29 |
+
"full_attention",
|
| 30 |
+
"full_attention",
|
| 31 |
+
"full_attention",
|
| 32 |
+
"full_attention",
|
| 33 |
+
"full_attention",
|
| 34 |
+
"full_attention",
|
| 35 |
+
"full_attention",
|
| 36 |
+
"full_attention",
|
| 37 |
+
"full_attention",
|
| 38 |
+
"full_attention",
|
| 39 |
+
"full_attention",
|
| 40 |
+
"full_attention",
|
| 41 |
+
"full_attention",
|
| 42 |
+
"full_attention",
|
| 43 |
+
"full_attention"
|
| 44 |
+
],
|
| 45 |
+
"max_position_embeddings": 40960,
|
| 46 |
+
"max_window_layers": 28,
|
| 47 |
+
"model_type": "qwen3",
|
| 48 |
+
"num_attention_heads": 16,
|
| 49 |
+
"num_hidden_layers": 28,
|
| 50 |
+
"num_key_value_heads": 8,
|
| 51 |
+
"rms_norm_eps": 1e-06,
|
| 52 |
+
"rope_scaling": null,
|
| 53 |
+
"rope_theta": 1000000,
|
| 54 |
+
"sliding_window": null,
|
| 55 |
+
"tie_word_embeddings": true,
|
| 56 |
+
"transformers_version": "4.57.3",
|
| 57 |
+
"use_cache": true,
|
| 58 |
+
"use_sliding_window": false,
|
| 59 |
+
"vocab_size": 151936
|
| 60 |
+
}
|
llm/generation_config.json
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 151643,
|
| 3 |
+
"do_sample": true,
|
| 4 |
+
"eos_token_id": [
|
| 5 |
+
151645,
|
| 6 |
+
151643
|
| 7 |
+
],
|
| 8 |
+
"lora_target": "qkvm",
|
| 9 |
+
"pad_token_id": 151643,
|
| 10 |
+
"temperature": 0.6,
|
| 11 |
+
"top_k": 20,
|
| 12 |
+
"top_p": 0.95,
|
| 13 |
+
"transformers_version": "4.57.3"
|
| 14 |
+
}
|
llm/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ca21804136ce2dcb829f3c11645cb182ed885eb333b527935ad77c9e90a12403
|
| 3 |
+
size 2384234968
|
lora_encoder.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:77f8b6c2666174793e7d206c811a90f80c00c3c12b42a81791005c0c66b27662
|
| 3 |
+
size 616399512
|
training_metadata.json
ADDED
|
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"global_step": 40000,
|
| 3 |
+
"epoch": 3,
|
| 4 |
+
"train_config": {
|
| 5 |
+
"lr": 1e-05,
|
| 6 |
+
"weight_decay": 1e-05,
|
| 7 |
+
"warmup_steps": 800,
|
| 8 |
+
"max_grad_norm": 1.0,
|
| 9 |
+
"train_llm": true,
|
| 10 |
+
"llm_lr": 1e-05,
|
| 11 |
+
"max_steps": 50000,
|
| 12 |
+
"log_every_n_steps": 10,
|
| 13 |
+
"val_every_n_steps": 5000,
|
| 14 |
+
"generate_every_n_steps": 5000,
|
| 15 |
+
"save_every_n_steps": 5000,
|
| 16 |
+
"num_samples_to_generate": 16,
|
| 17 |
+
"generation_max_new_tokens": 128,
|
| 18 |
+
"generation_temperature": 0.4,
|
| 19 |
+
"generation_do_sample": true,
|
| 20 |
+
"compute_loss_on_assistant_only": true,
|
| 21 |
+
"wandb_project": "smol-loravlm",
|
| 22 |
+
"wandb_run_name": "base-ft-continue-3k"
|
| 23 |
+
},
|
| 24 |
+
"checkpoint_config": {
|
| 25 |
+
"checkpoint_dir": "checkpoints/smol_loravlm_base_ft_from_3000",
|
| 26 |
+
"push_to_hub": false,
|
| 27 |
+
"hub_model_id": "toilaluan/ai0",
|
| 28 |
+
"hub_token": null,
|
| 29 |
+
"hub_private_repo": true
|
| 30 |
+
},
|
| 31 |
+
"model_config": {
|
| 32 |
+
"text_ckpt": "google/embeddinggemma-300m",
|
| 33 |
+
"image_ckpt": "google/siglip2-so400m-patch16-256",
|
| 34 |
+
"llm_ckpt": "checkpoints/smol_loravlm_base_ft/checkpoint_step_3000/llm",
|
| 35 |
+
"lora_rank": 16,
|
| 36 |
+
"resume_from_checkpoint": "checkpoints/smol_loravlm_base_ft/checkpoint_step_3000",
|
| 37 |
+
"lora_target": "qkvm",
|
| 38 |
+
"prompt_aware": false,
|
| 39 |
+
"lora_use_prefix": true
|
| 40 |
+
}
|
| 41 |
+
}
|