Duplicate from zhengyuansu/bagel-tiny-random

Co-authored-by: Zhengyuan Su <zhengyuansu@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,196 @@
+---
+library_name: vllm-omni
+pipeline_tag: text-to-image
+inference: true
+base_model:
+- ByteDance-Seed/BAGEL-7B-MoT
+---
+
+This tiny model is for debugging. It is randomly initialized with the config adapted from [ByteDance-Seed/BAGEL-7B-MoT](https://huggingface.co/ByteDance-Seed/BAGEL-7B-MoT).
+
+File size:
+- ~335MB ae.safetensors (VAE, full architecture — hardcoded in vllm-omni)
+- ~41MB  ema.safetensors (LLM + ViT + connectors, 1 layer each)
+
+### Example usage:
+
+```python
+from vllm_omni.entrypoints.omni import Omni
+
+omni = Omni(
+    model="zhengyuansu/bagel-tiny-random",
+    stage_configs_path="path/to/bagel_sharedmemory_2gpu_ci.yaml",
+    custom_pipeline_args={
+        "pipeline_class": "examples.flowgrpo_trainer.vllm_omni.pipeline_bagel.BagelPipelineWithLogProb"
+    },
+)
+
+params_list = omni.default_sampling_params_list
+params_list[1].num_inference_steps = 10
+params_list[1].extra_args = {"cfg_text_scale": 4.0, "cfg_img_scale": 1.5}
+
+outputs = list(omni.generate(
+    prompts=[{"prompt": "a cute cat", "modalities": ["image"]}],
+    sampling_params_list=params_list,
+))
+```
+
+### Codes to create this repo:
+
+```python
+"""Create a tiny-random BAGEL model for CI testing.
+
+Reads real BAGEL-7B-MoT checkpoint weight names, creates matching tiny random
+tensors with scaled-down dimensions. VAE architecture is hardcoded in vllm-omni
+and cannot be shrunk, so VAE weights are kept at full size.
+
+Usage:
+    python scripts/create_tiny_bagel.py --source ByteDance-Seed/BAGEL-7B-MoT
+"""
+
+import argparse
+import json
+import os
+import re
+import shutil
+
+import torch
+from safetensors import safe_open
+from safetensors.torch import save_file
+
+# LLM/ViT dimension shrinkage
+EMA_DIM_MAP = {
+    3584: 64,     # LLM hidden_size
+    18944: 128,   # LLM intermediate_size
+    1152: 64,     # ViT hidden_size
+    4304: 128,    # ViT intermediate_size
+    128: 32,      # head_dim
+    512: 64,      # kv_proj dim
+}
+
+# VAE: keep original dims (architecture is hardcoded in vllm-omni)
+VAE_DIM_MAP = {}
+
+MAX_LLM_LAYERS = 1
+MAX_VIT_LAYERS = 1
+
+
+def shrink_dims(shape, dim_map):
+    return [dim_map.get(d, d) for d in shape]
+
+
+def create_tiny_configs(source_dir, output_dir):
+    with open(os.path.join(source_dir, "config.json")) as f:
+        config = json.load(f)
+
+    llm = config["llm_config"]
+    llm["hidden_size"] = 64
+    llm["num_hidden_layers"] = MAX_LLM_LAYERS
+    llm["num_attention_heads"] = 2
+    llm["num_key_value_heads"] = 2
+    llm["intermediate_size"] = 128
+    llm["max_position_embeddings"] = 4096
+    llm["max_window_layers"] = MAX_LLM_LAYERS
+
+    vit = config["vit_config"]
+    vit["hidden_size"] = 64
+    vit["num_hidden_layers"] = MAX_VIT_LAYERS
+    vit["num_attention_heads"] = 2
+    vit["intermediate_size"] = 128
+
+    with open(os.path.join(output_dir, "config.json"), "w") as f:
+        json.dump(config, f, indent=4)
+
+    llm_standalone = dict(llm)
+    llm_standalone["qk_norm"] = True
+    llm_standalone["tie_word_embeddings"] = False
+    with open(os.path.join(output_dir, "llm_config.json"), "w") as f:
+        json.dump(llm_standalone, f, indent=4)
+
+    with open(os.path.join(output_dir, "vit_config.json"), "w") as f:
+        json.dump(dict(vit), f, indent=4)
+
+    return config
+
+
+def create_tiny_weights(source_path, dim_map, max_layers, seed=42):
+    gen = torch.Generator().manual_seed(seed)
+    weights = {}
+    is_vae = "ae" in os.path.basename(source_path).lower()
+    dtype = torch.float32 if is_vae else torch.bfloat16
+
+    with safe_open(source_path, framework="pt") as f:
+        for name in f.keys():
+            m = re.search(r"\.layers\.(\d+)\.", name)
+            if m:
+                idx = int(m.group(1))
+                for pattern, limit in max_layers.items():
+                    if pattern in name and idx >= limit:
+                        break
+                else:
+                    pass
+                if m and any(p in name for p in max_layers) and idx >= max_layers.get(
+                    next((p for p in max_layers if p in name), ""), 999
+                ):
+                    continue
+
+            real_shape = list(f.get_tensor(name).shape)
+            tiny_shape = shrink_dims(real_shape, dim_map)
+
+            if "norm" in name and len(tiny_shape) == 1:
+                weights[name] = torch.ones(tiny_shape, dtype=dtype)
+            else:
+                weights[name] = torch.randn(tiny_shape, generator=gen, dtype=dtype) * 0.02
+
+    return weights
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--source", default="ByteDance-Seed/BAGEL-7B-MoT")
+    parser.add_argument("--output", default=os.path.expanduser("~/models/tiny-random/BAGEL-7B-MoT"))
+    args = parser.parse_args()
+
+    source_dir = args.source
+    if not os.path.exists(os.path.join(source_dir, "config.json")):
+        from huggingface_hub import snapshot_download
+        source_dir = snapshot_download(source_dir)
+
+    output_dir = args.output
+    os.makedirs(output_dir, exist_ok=True)
+
+    create_tiny_configs(source_dir, output_dir)
+
+    for fname in ["generation_config.json", "preprocessor_config.json", "tokenizer.json",
+                   "tokenizer_config.json", "vocab.json", "merges.txt"]:
+        src = os.path.join(source_dir, fname)
+        if os.path.exists(src):
+            shutil.copy2(src, os.path.join(output_dir, fname))
+
+    ema = create_tiny_weights(
+        os.path.join(source_dir, "ema.safetensors"),
+        dim_map=EMA_DIM_MAP,
+        max_layers={"language_model": MAX_LLM_LAYERS, "vit_model": MAX_VIT_LAYERS},
+        seed=42,
+    )
+    save_file(ema, os.path.join(output_dir, "ema.safetensors"))
+
+    vae = create_tiny_weights(
+        os.path.join(source_dir, "ae.safetensors"),
+        dim_map=VAE_DIM_MAP,
+        max_layers={},
+        seed=43,
+    )
+    save_file(vae, os.path.join(output_dir, "ae.safetensors"))
+
+    weight_map = {k: "ema.safetensors" for k in ema}
+    weight_map.update({k: "ae.safetensors" for k in vae})
+    total_size = sum(t.numel() * t.element_size() for t in ema.values())
+    total_size += sum(t.numel() * t.element_size() for t in vae.values())
+    with open(os.path.join(output_dir, "model.safetensors.index.json"), "w") as f:
+        json.dump({"metadata": {"total_size": total_size}, "weight_map": weight_map}, f, indent=4)
+
+
+if __name__ == "__main__":
+    main()
+```

ae.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a5fa0e75c982c563fbf8c188b953f9307749f7f7e3a9e1ff68cd6a284fcbc2a
+size 335303916

config.json

@@ -0,0 +1,58 @@
+{
+    "architectures": [
+        "BagelForConditionalGeneration"
+    ],
+    "model_type": "bagel",
+    "visual_gen": true,
+    "visual_und": true,
+    "llm_config": {
+        "architectures": [
+            "Qwen2ForCausalLM"
+        ],
+        "attention_dropout": 0.0,
+        "bos_token_id": 151643,
+        "eos_token_id": 151645,
+        "hidden_act": "silu",
+        "hidden_size": 64,
+        "initializer_range": 0.02,
+        "intermediate_size": 128,
+        "max_position_embeddings": 4096,
+        "max_window_layers": 1,
+        "model_type": "qwen2",
+        "num_attention_heads": 2,
+        "num_hidden_layers": 1,
+        "num_key_value_heads": 2,
+        "qk_norm": true,
+        "rms_norm_eps": 1e-06,
+        "rope_theta": 1000000.0,
+        "sliding_window": 131072,
+        "tie_word_embeddings": false,
+        "torch_dtype": "bfloat16",
+        "transformers_version": "4.43.1",
+        "use_cache": true,
+        "use_sliding_window": false,
+        "vocab_size": 152064
+    },
+    "vit_config": {
+        "hidden_size": 64,
+        "image_size": 980,
+        "intermediate_size": 128,
+        "model_type": "siglip_vision_model",
+        "num_attention_heads": 2,
+        "num_hidden_layers": 1,
+        "patch_size": 14,
+        "num_channels": 3
+    },
+    "vae_config": {
+        "z_channels": 16,
+        "downsample": 8
+    },
+    "latent_patch_size": 2,
+    "max_latent_size": 32,
+    "vit_max_num_patch_per_side": 70,
+    "connector_act": "gelu_pytorch_tanh",
+    "interpolate_pos": false,
+    "timestep_shift": 1.0,
+    "torch_dtype": "bfloat16",
+    "transformers_version": "4.49.0"
+}

ema.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cff4c7b68f95b4abaf0fdeaa6593b9bb3f33b61623dfad14759b2d80ca402da
+size 41097632

generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "pad_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "repetition_penalty": 1.05,
+  "temperature": 0.7,
+  "top_p": 0.8,
+  "top_k": 20,
+  "transformers_version": "4.37.0"
+}

llm_config.json

@@ -0,0 +1,28 @@
+{
+    "architectures": [
+        "Qwen2ForCausalLM"
+    ],
+    "attention_dropout": 0.0,
+    "bos_token_id": 151643,
+    "eos_token_id": 151645,
+    "hidden_act": "silu",
+    "hidden_size": 64,
+    "initializer_range": 0.02,
+    "intermediate_size": 128,
+    "max_position_embeddings": 4096,
+    "max_window_layers": 1,
+    "model_type": "qwen2",
+    "num_attention_heads": 2,
+    "num_hidden_layers": 1,
+    "num_key_value_heads": 2,
+    "qk_norm": true,
+    "rms_norm_eps": 1e-06,
+    "rope_theta": 1000000.0,
+    "sliding_window": 131072,
+    "tie_word_embeddings": false,
+    "torch_dtype": "bfloat16",
+    "transformers_version": "4.43.1",
+    "use_cache": true,
+    "use_sliding_window": false,
+    "vocab_size": 152064
+}

model.safetensors.index.json

@@ -0,0 +1,318 @@
+{
+    "metadata": {
+        "total_size": 376368524
+    },
+    "weight_map": {
+        "connector.fc1.bias": "ema.safetensors",
+        "connector.fc1.weight": "ema.safetensors",
+        "connector.fc2.bias": "ema.safetensors",
+        "connector.fc2.weight": "ema.safetensors",
+        "language_model.lm_head.weight": "ema.safetensors",
+        "language_model.model.embed_tokens.weight": "ema.safetensors",
+        "language_model.model.layers.0.input_layernorm.weight": "ema.safetensors",
+        "language_model.model.layers.0.input_layernorm_moe_gen.weight": "ema.safetensors",
+        "language_model.model.layers.0.mlp.down_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.mlp.gate_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.mlp.up_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.mlp_moe_gen.down_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.mlp_moe_gen.gate_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.mlp_moe_gen.up_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.post_attention_layernorm.weight": "ema.safetensors",
+        "language_model.model.layers.0.post_attention_layernorm_moe_gen.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.k_norm.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.k_norm_moe_gen.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.k_proj.bias": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.k_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.k_proj_moe_gen.bias": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.k_proj_moe_gen.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.o_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.o_proj_moe_gen.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.q_norm.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.q_norm_moe_gen.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.q_proj.bias": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.q_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.q_proj_moe_gen.bias": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.q_proj_moe_gen.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.v_proj.bias": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.v_proj.weight": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.v_proj_moe_gen.bias": "ema.safetensors",
+        "language_model.model.layers.0.self_attn.v_proj_moe_gen.weight": "ema.safetensors",
+        "language_model.model.norm.weight": "ema.safetensors",
+        "language_model.model.norm_moe_gen.weight": "ema.safetensors",
+        "latent_pos_embed.pos_embed": "ema.safetensors",
+        "llm2vae.bias": "ema.safetensors",
+        "llm2vae.weight": "ema.safetensors",
+        "time_embedder.mlp.0.bias": "ema.safetensors",
+        "time_embedder.mlp.0.weight": "ema.safetensors",
+        "time_embedder.mlp.2.bias": "ema.safetensors",
+        "time_embedder.mlp.2.weight": "ema.safetensors",
+        "vae2llm.bias": "ema.safetensors",
+        "vae2llm.weight": "ema.safetensors",
+        "vit_model.vision_model.embeddings.patch_embedding.bias": "ema.safetensors",
+        "vit_model.vision_model.embeddings.patch_embedding.weight": "ema.safetensors",
+        "vit_model.vision_model.embeddings.position_embedding.weight": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.layer_norm1.bias": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.layer_norm1.weight": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.layer_norm2.bias": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.layer_norm2.weight": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.mlp.fc1.bias": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.mlp.fc1.weight": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.mlp.fc2.bias": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.mlp.fc2.weight": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.self_attn.k_proj.bias": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.self_attn.k_proj.weight": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.self_attn.out_proj.bias": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.self_attn.out_proj.weight": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.self_attn.q_proj.bias": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.self_attn.q_proj.weight": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.self_attn.v_proj.bias": "ema.safetensors",
+        "vit_model.vision_model.encoder.layers.0.self_attn.v_proj.weight": "ema.safetensors",
+        "vit_model.vision_model.post_layernorm.bias": "ema.safetensors",
+        "vit_model.vision_model.post_layernorm.weight": "ema.safetensors",
+        "vit_pos_embed.pos_embed": "ema.safetensors",
+        "decoder.conv_in.bias": "ae.safetensors",
+        "decoder.conv_in.weight": "ae.safetensors",
+        "decoder.conv_out.bias": "ae.safetensors",
+        "decoder.conv_out.weight": "ae.safetensors",
+        "decoder.mid.attn_1.k.bias": "ae.safetensors",
+        "decoder.mid.attn_1.k.weight": "ae.safetensors",
+        "decoder.mid.attn_1.norm.bias": "ae.safetensors",
+        "decoder.mid.attn_1.norm.weight": "ae.safetensors",
+        "decoder.mid.attn_1.proj_out.bias": "ae.safetensors",
+        "decoder.mid.attn_1.proj_out.weight": "ae.safetensors",
+        "decoder.mid.attn_1.q.bias": "ae.safetensors",
+        "decoder.mid.attn_1.q.weight": "ae.safetensors",
+        "decoder.mid.attn_1.v.bias": "ae.safetensors",
+        "decoder.mid.attn_1.v.weight": "ae.safetensors",
+        "decoder.mid.block_1.conv1.bias": "ae.safetensors",
+        "decoder.mid.block_1.conv1.weight": "ae.safetensors",
+        "decoder.mid.block_1.conv2.bias": "ae.safetensors",
+        "decoder.mid.block_1.conv2.weight": "ae.safetensors",
+        "decoder.mid.block_1.norm1.bias": "ae.safetensors",
+        "decoder.mid.block_1.norm1.weight": "ae.safetensors",
+        "decoder.mid.block_1.norm2.bias": "ae.safetensors",
+        "decoder.mid.block_1.norm2.weight": "ae.safetensors",
+        "decoder.mid.block_2.conv1.bias": "ae.safetensors",
+        "decoder.mid.block_2.conv1.weight": "ae.safetensors",
+        "decoder.mid.block_2.conv2.bias": "ae.safetensors",
+        "decoder.mid.block_2.conv2.weight": "ae.safetensors",
+        "decoder.mid.block_2.norm1.bias": "ae.safetensors",
+        "decoder.mid.block_2.norm1.weight": "ae.safetensors",
+        "decoder.mid.block_2.norm2.bias": "ae.safetensors",
+        "decoder.mid.block_2.norm2.weight": "ae.safetensors",
+        "decoder.norm_out.bias": "ae.safetensors",
+        "decoder.norm_out.weight": "ae.safetensors",
+        "decoder.up.0.block.0.conv1.bias": "ae.safetensors",
+        "decoder.up.0.block.0.conv1.weight": "ae.safetensors",
+        "decoder.up.0.block.0.conv2.bias": "ae.safetensors",
+        "decoder.up.0.block.0.conv2.weight": "ae.safetensors",
+        "decoder.up.0.block.0.nin_shortcut.bias": "ae.safetensors",
+        "decoder.up.0.block.0.nin_shortcut.weight": "ae.safetensors",
+        "decoder.up.0.block.0.norm1.bias": "ae.safetensors",
+        "decoder.up.0.block.0.norm1.weight": "ae.safetensors",
+        "decoder.up.0.block.0.norm2.bias": "ae.safetensors",
+        "decoder.up.0.block.0.norm2.weight": "ae.safetensors",
+        "decoder.up.0.block.1.conv1.bias": "ae.safetensors",
+        "decoder.up.0.block.1.conv1.weight": "ae.safetensors",
+        "decoder.up.0.block.1.conv2.bias": "ae.safetensors",
+        "decoder.up.0.block.1.conv2.weight": "ae.safetensors",
+        "decoder.up.0.block.1.norm1.bias": "ae.safetensors",
+        "decoder.up.0.block.1.norm1.weight": "ae.safetensors",
+        "decoder.up.0.block.1.norm2.bias": "ae.safetensors",
+        "decoder.up.0.block.1.norm2.weight": "ae.safetensors",
+        "decoder.up.0.block.2.conv1.bias": "ae.safetensors",
+        "decoder.up.0.block.2.conv1.weight": "ae.safetensors",
+        "decoder.up.0.block.2.conv2.bias": "ae.safetensors",
+        "decoder.up.0.block.2.conv2.weight": "ae.safetensors",
+        "decoder.up.0.block.2.norm1.bias": "ae.safetensors",
+        "decoder.up.0.block.2.norm1.weight": "ae.safetensors",
+        "decoder.up.0.block.2.norm2.bias": "ae.safetensors",
+        "decoder.up.0.block.2.norm2.weight": "ae.safetensors",
+        "decoder.up.1.block.0.conv1.bias": "ae.safetensors",
+        "decoder.up.1.block.0.conv1.weight": "ae.safetensors",
+        "decoder.up.1.block.0.conv2.bias": "ae.safetensors",
+        "decoder.up.1.block.0.conv2.weight": "ae.safetensors",
+        "decoder.up.1.block.0.nin_shortcut.bias": "ae.safetensors",
+        "decoder.up.1.block.0.nin_shortcut.weight": "ae.safetensors",
+        "decoder.up.1.block.0.norm1.bias": "ae.safetensors",
+        "decoder.up.1.block.0.norm1.weight": "ae.safetensors",
+        "decoder.up.1.block.0.norm2.bias": "ae.safetensors",
+        "decoder.up.1.block.0.norm2.weight": "ae.safetensors",
+        "decoder.up.1.block.1.conv1.bias": "ae.safetensors",
+        "decoder.up.1.block.1.conv1.weight": "ae.safetensors",
+        "decoder.up.1.block.1.conv2.bias": "ae.safetensors",
+        "decoder.up.1.block.1.conv2.weight": "ae.safetensors",
+        "decoder.up.1.block.1.norm1.bias": "ae.safetensors",
+        "decoder.up.1.block.1.norm1.weight": "ae.safetensors",
+        "decoder.up.1.block.1.norm2.bias": "ae.safetensors",
+        "decoder.up.1.block.1.norm2.weight": "ae.safetensors",
+        "decoder.up.1.block.2.conv1.bias": "ae.safetensors",
+        "decoder.up.1.block.2.conv1.weight": "ae.safetensors",
+        "decoder.up.1.block.2.conv2.bias": "ae.safetensors",
+        "decoder.up.1.block.2.conv2.weight": "ae.safetensors",
+        "decoder.up.1.block.2.norm1.bias": "ae.safetensors",
+        "decoder.up.1.block.2.norm1.weight": "ae.safetensors",
+        "decoder.up.1.block.2.norm2.bias": "ae.safetensors",
+        "decoder.up.1.block.2.norm2.weight": "ae.safetensors",
+        "decoder.up.1.upsample.conv.bias": "ae.safetensors",
+        "decoder.up.1.upsample.conv.weight": "ae.safetensors",
+        "decoder.up.2.block.0.conv1.bias": "ae.safetensors",
+        "decoder.up.2.block.0.conv1.weight": "ae.safetensors",
+        "decoder.up.2.block.0.conv2.bias": "ae.safetensors",
+        "decoder.up.2.block.0.conv2.weight": "ae.safetensors",
+        "decoder.up.2.block.0.norm1.bias": "ae.safetensors",
+        "decoder.up.2.block.0.norm1.weight": "ae.safetensors",
+        "decoder.up.2.block.0.norm2.bias": "ae.safetensors",
+        "decoder.up.2.block.0.norm2.weight": "ae.safetensors",
+        "decoder.up.2.block.1.conv1.bias": "ae.safetensors",
+        "decoder.up.2.block.1.conv1.weight": "ae.safetensors",
+        "decoder.up.2.block.1.conv2.bias": "ae.safetensors",
+        "decoder.up.2.block.1.conv2.weight": "ae.safetensors",
+        "decoder.up.2.block.1.norm1.bias": "ae.safetensors",
+        "decoder.up.2.block.1.norm1.weight": "ae.safetensors",
+        "decoder.up.2.block.1.norm2.bias": "ae.safetensors",
+        "decoder.up.2.block.1.norm2.weight": "ae.safetensors",
+        "decoder.up.2.block.2.conv1.bias": "ae.safetensors",
+        "decoder.up.2.block.2.conv1.weight": "ae.safetensors",
+        "decoder.up.2.block.2.conv2.bias": "ae.safetensors",
+        "decoder.up.2.block.2.conv2.weight": "ae.safetensors",
+        "decoder.up.2.block.2.norm1.bias": "ae.safetensors",
+        "decoder.up.2.block.2.norm1.weight": "ae.safetensors",
+        "decoder.up.2.block.2.norm2.bias": "ae.safetensors",
+        "decoder.up.2.block.2.norm2.weight": "ae.safetensors",
+        "decoder.up.2.upsample.conv.bias": "ae.safetensors",
+        "decoder.up.2.upsample.conv.weight": "ae.safetensors",
+        "decoder.up.3.block.0.conv1.bias": "ae.safetensors",
+        "decoder.up.3.block.0.conv1.weight": "ae.safetensors",
+        "decoder.up.3.block.0.conv2.bias": "ae.safetensors",
+        "decoder.up.3.block.0.conv2.weight": "ae.safetensors",
+        "decoder.up.3.block.0.norm1.bias": "ae.safetensors",
+        "decoder.up.3.block.0.norm1.weight": "ae.safetensors",
+        "decoder.up.3.block.0.norm2.bias": "ae.safetensors",
+        "decoder.up.3.block.0.norm2.weight": "ae.safetensors",
+        "decoder.up.3.block.1.conv1.bias": "ae.safetensors",
+        "decoder.up.3.block.1.conv1.weight": "ae.safetensors",
+        "decoder.up.3.block.1.conv2.bias": "ae.safetensors",
+        "decoder.up.3.block.1.conv2.weight": "ae.safetensors",
+        "decoder.up.3.block.1.norm1.bias": "ae.safetensors",
+        "decoder.up.3.block.1.norm1.weight": "ae.safetensors",
+        "decoder.up.3.block.1.norm2.bias": "ae.safetensors",
+        "decoder.up.3.block.1.norm2.weight": "ae.safetensors",
+        "decoder.up.3.block.2.conv1.bias": "ae.safetensors",
+        "decoder.up.3.block.2.conv1.weight": "ae.safetensors",
+        "decoder.up.3.block.2.conv2.bias": "ae.safetensors",
+        "decoder.up.3.block.2.conv2.weight": "ae.safetensors",
+        "decoder.up.3.block.2.norm1.bias": "ae.safetensors",
+        "decoder.up.3.block.2.norm1.weight": "ae.safetensors",
+        "decoder.up.3.block.2.norm2.bias": "ae.safetensors",
+        "decoder.up.3.block.2.norm2.weight": "ae.safetensors",
+        "decoder.up.3.upsample.conv.bias": "ae.safetensors",
+        "decoder.up.3.upsample.conv.weight": "ae.safetensors",
+        "encoder.conv_in.bias": "ae.safetensors",
+        "encoder.conv_in.weight": "ae.safetensors",
+        "encoder.conv_out.bias": "ae.safetensors",
+        "encoder.conv_out.weight": "ae.safetensors",
+        "encoder.down.0.block.0.conv1.bias": "ae.safetensors",
+        "encoder.down.0.block.0.conv1.weight": "ae.safetensors",
+        "encoder.down.0.block.0.conv2.bias": "ae.safetensors",
+        "encoder.down.0.block.0.conv2.weight": "ae.safetensors",
+        "encoder.down.0.block.0.norm1.bias": "ae.safetensors",
+        "encoder.down.0.block.0.norm1.weight": "ae.safetensors",
+        "encoder.down.0.block.0.norm2.bias": "ae.safetensors",
+        "encoder.down.0.block.0.norm2.weight": "ae.safetensors",
+        "encoder.down.0.block.1.conv1.bias": "ae.safetensors",
+        "encoder.down.0.block.1.conv1.weight": "ae.safetensors",
+        "encoder.down.0.block.1.conv2.bias": "ae.safetensors",
+        "encoder.down.0.block.1.conv2.weight": "ae.safetensors",
+        "encoder.down.0.block.1.norm1.bias": "ae.safetensors",
+        "encoder.down.0.block.1.norm1.weight": "ae.safetensors",
+        "encoder.down.0.block.1.norm2.bias": "ae.safetensors",
+        "encoder.down.0.block.1.norm2.weight": "ae.safetensors",
+        "encoder.down.0.downsample.conv.bias": "ae.safetensors",
+        "encoder.down.0.downsample.conv.weight": "ae.safetensors",
+        "encoder.down.1.block.0.conv1.bias": "ae.safetensors",
+        "encoder.down.1.block.0.conv1.weight": "ae.safetensors",
+        "encoder.down.1.block.0.conv2.bias": "ae.safetensors",
+        "encoder.down.1.block.0.conv2.weight": "ae.safetensors",
+        "encoder.down.1.block.0.nin_shortcut.bias": "ae.safetensors",
+        "encoder.down.1.block.0.nin_shortcut.weight": "ae.safetensors",
+        "encoder.down.1.block.0.norm1.bias": "ae.safetensors",
+        "encoder.down.1.block.0.norm1.weight": "ae.safetensors",
+        "encoder.down.1.block.0.norm2.bias": "ae.safetensors",
+        "encoder.down.1.block.0.norm2.weight": "ae.safetensors",
+        "encoder.down.1.block.1.conv1.bias": "ae.safetensors",
+        "encoder.down.1.block.1.conv1.weight": "ae.safetensors",
+        "encoder.down.1.block.1.conv2.bias": "ae.safetensors",
+        "encoder.down.1.block.1.conv2.weight": "ae.safetensors",
+        "encoder.down.1.block.1.norm1.bias": "ae.safetensors",
+        "encoder.down.1.block.1.norm1.weight": "ae.safetensors",
+        "encoder.down.1.block.1.norm2.bias": "ae.safetensors",
+        "encoder.down.1.block.1.norm2.weight": "ae.safetensors",
+        "encoder.down.1.downsample.conv.bias": "ae.safetensors",
+        "encoder.down.1.downsample.conv.weight": "ae.safetensors",
+        "encoder.down.2.block.0.conv1.bias": "ae.safetensors",
+        "encoder.down.2.block.0.conv1.weight": "ae.safetensors",
+        "encoder.down.2.block.0.conv2.bias": "ae.safetensors",
+        "encoder.down.2.block.0.conv2.weight": "ae.safetensors",
+        "encoder.down.2.block.0.nin_shortcut.bias": "ae.safetensors",
+        "encoder.down.2.block.0.nin_shortcut.weight": "ae.safetensors",
+        "encoder.down.2.block.0.norm1.bias": "ae.safetensors",
+        "encoder.down.2.block.0.norm1.weight": "ae.safetensors",
+        "encoder.down.2.block.0.norm2.bias": "ae.safetensors",
+        "encoder.down.2.block.0.norm2.weight": "ae.safetensors",
+        "encoder.down.2.block.1.conv1.bias": "ae.safetensors",
+        "encoder.down.2.block.1.conv1.weight": "ae.safetensors",
+        "encoder.down.2.block.1.conv2.bias": "ae.safetensors",
+        "encoder.down.2.block.1.conv2.weight": "ae.safetensors",
+        "encoder.down.2.block.1.norm1.bias": "ae.safetensors",
+        "encoder.down.2.block.1.norm1.weight": "ae.safetensors",
+        "encoder.down.2.block.1.norm2.bias": "ae.safetensors",
+        "encoder.down.2.block.1.norm2.weight": "ae.safetensors",
+        "encoder.down.2.downsample.conv.bias": "ae.safetensors",
+        "encoder.down.2.downsample.conv.weight": "ae.safetensors",
+        "encoder.down.3.block.0.conv1.bias": "ae.safetensors",
+        "encoder.down.3.block.0.conv1.weight": "ae.safetensors",
+        "encoder.down.3.block.0.conv2.bias": "ae.safetensors",
+        "encoder.down.3.block.0.conv2.weight": "ae.safetensors",
+        "encoder.down.3.block.0.norm1.bias": "ae.safetensors",
+        "encoder.down.3.block.0.norm1.weight": "ae.safetensors",
+        "encoder.down.3.block.0.norm2.bias": "ae.safetensors",
+        "encoder.down.3.block.0.norm2.weight": "ae.safetensors",
+        "encoder.down.3.block.1.conv1.bias": "ae.safetensors",
+        "encoder.down.3.block.1.conv1.weight": "ae.safetensors",
+        "encoder.down.3.block.1.conv2.bias": "ae.safetensors",
+        "encoder.down.3.block.1.conv2.weight": "ae.safetensors",
+        "encoder.down.3.block.1.norm1.bias": "ae.safetensors",
+        "encoder.down.3.block.1.norm1.weight": "ae.safetensors",
+        "encoder.down.3.block.1.norm2.bias": "ae.safetensors",
+        "encoder.down.3.block.1.norm2.weight": "ae.safetensors",
+        "encoder.mid.attn_1.k.bias": "ae.safetensors",
+        "encoder.mid.attn_1.k.weight": "ae.safetensors",
+        "encoder.mid.attn_1.norm.bias": "ae.safetensors",
+        "encoder.mid.attn_1.norm.weight": "ae.safetensors",
+        "encoder.mid.attn_1.proj_out.bias": "ae.safetensors",
+        "encoder.mid.attn_1.proj_out.weight": "ae.safetensors",
+        "encoder.mid.attn_1.q.bias": "ae.safetensors",
+        "encoder.mid.attn_1.q.weight": "ae.safetensors",
+        "encoder.mid.attn_1.v.bias": "ae.safetensors",
+        "encoder.mid.attn_1.v.weight": "ae.safetensors",
+        "encoder.mid.block_1.conv1.bias": "ae.safetensors",
+        "encoder.mid.block_1.conv1.weight": "ae.safetensors",
+        "encoder.mid.block_1.conv2.bias": "ae.safetensors",
+        "encoder.mid.block_1.conv2.weight": "ae.safetensors",
+        "encoder.mid.block_1.norm1.bias": "ae.safetensors",
+        "encoder.mid.block_1.norm1.weight": "ae.safetensors",
+        "encoder.mid.block_1.norm2.bias": "ae.safetensors",
+        "encoder.mid.block_1.norm2.weight": "ae.safetensors",
+        "encoder.mid.block_2.conv1.bias": "ae.safetensors",
+        "encoder.mid.block_2.conv1.weight": "ae.safetensors",
+        "encoder.mid.block_2.conv2.bias": "ae.safetensors",
+        "encoder.mid.block_2.conv2.weight": "ae.safetensors",
+        "encoder.mid.block_2.norm1.bias": "ae.safetensors",
+        "encoder.mid.block_2.norm1.weight": "ae.safetensors",
+        "encoder.mid.block_2.norm2.bias": "ae.safetensors",
+        "encoder.mid.block_2.norm2.weight": "ae.safetensors",
+        "encoder.norm_out.bias": "ae.safetensors",
+        "encoder.norm_out.weight": "ae.safetensors"
+    }
+}

preprocessor_config.json

@@ -0,0 +1,24 @@
+{
+  "do_convert_rgb": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_mean": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "image_processor_type": "SiglipImageProcessor",
+  "image_std": [
+    0.5,
+    0.5,
+    0.5
+  ],
+  "processor_class": "BagelProcessor",
+  "rescale_factor": 0.00392156862745098,
+  "resample": 3,
+  "size": {
+    "height": 980,
+    "width": 980
+  }
+}

tokenizer_config.json

@@ -0,0 +1,207 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}\n    {%- endif %}\n    {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

vit_config.json

@@ -0,0 +1,10 @@
+{
+    "hidden_size": 64,
+    "image_size": 980,
+    "intermediate_size": 128,
+    "model_type": "siglip_vision_model",
+    "num_attention_heads": 2,
+    "num_hidden_layers": 1,
+    "patch_size": 14,
+    "num_channels": 3
+}