Image-Text-to-Text
Transformers
Safetensors
inkling2
conversational
Mixture of Experts
8-bit precision
Instructions to use thinkingmachines/Inkling-2-Small with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use thinkingmachines/Inkling-2-Small with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="thinkingmachines/Inkling-2-Small") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("thinkingmachines/Inkling-2-Small") model = AutoModelForMultimodalLM.from_pretrained("thinkingmachines/Inkling-2-Small", device_map="auto") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Inference
- HuggingChat
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use thinkingmachines/Inkling-2-Small with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "thinkingmachines/Inkling-2-Small" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "thinkingmachines/Inkling-2-Small", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker
docker model run hf.co/thinkingmachines/Inkling-2-Small
- SGLang
How to use thinkingmachines/Inkling-2-Small with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "thinkingmachines/Inkling-2-Small" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "thinkingmachines/Inkling-2-Small", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "thinkingmachines/Inkling-2-Small" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "thinkingmachines/Inkling-2-Small", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }' - Docker Model Runner
How to use thinkingmachines/Inkling-2-Small with Docker Model Runner:
docker model run hf.co/thinkingmachines/Inkling-2-Small
Download hf_quant_config.json from thinkingmachines/Inkling-2-Small: direct link, hf CLI and curl.
- Browser
- Download file 88.9 kB
-
https://huggingface.co/thinkingmachines/Inkling-2-Small/resolve/refs%2Fpr%2F1/hf_quant_config.json
- Command line
-
hf download hf://thinkingmachines/Inkling-2-Small@refs/pr/1/hf_quant_config.json
-
curl -L -o hf_quant_config.json https://huggingface.co/thinkingmachines/Inkling-2-Small/resolve/refs%2Fpr%2F1/hf_quant_config.json
88.9 kB
{
"model": {
"parameter_groups": [
{
"compute_dtype": "float32",
"output_dtype": "float32",
"required_activation_quantization": null,
"tensors": [
"model.llm.attn_res.attn_queries.0",
"model.llm.attn_res.attn_queries.1",
"model.llm.attn_res.attn_queries.10",
"model.llm.attn_res.attn_queries.11",
"model.llm.attn_res.attn_queries.12",
"model.llm.attn_res.attn_queries.13",
"model.llm.attn_res.attn_queries.14",
"model.llm.attn_res.attn_queries.15",
"model.llm.attn_res.attn_queries.16",
"model.llm.attn_res.attn_queries.17",
"model.llm.attn_res.attn_queries.18",
"model.llm.attn_res.attn_queries.19",
"model.llm.attn_res.attn_queries.2",
"model.llm.attn_res.attn_queries.20",
"model.llm.attn_res.attn_queries.21",
"model.llm.attn_res.attn_queries.22",
"model.llm.attn_res.attn_queries.23",
"model.llm.attn_res.attn_queries.24",
"model.llm.attn_res.attn_queries.25",
"model.llm.attn_res.attn_queries.26",
"model.llm.attn_res.attn_queries.27",
"model.llm.attn_res.attn_queries.28",
"model.llm.attn_res.attn_queries.29",
"model.llm.attn_res.attn_queries.3",
"model.llm.attn_res.attn_queries.30",
"model.llm.attn_res.attn_queries.31",
"model.llm.attn_res.attn_queries.32",
"model.llm.attn_res.attn_queries.33",
"model.llm.attn_res.attn_queries.34",
"model.llm.attn_res.attn_queries.35",
"model.llm.attn_res.attn_queries.36",
"model.llm.attn_res.attn_queries.37",
"model.llm.attn_res.attn_queries.38",
"model.llm.attn_res.attn_queries.39",
"model.llm.attn_res.attn_queries.4",
"model.llm.attn_res.attn_queries.40",
"model.llm.attn_res.attn_queries.41",
"model.llm.attn_res.attn_queries.42",
"model.llm.attn_res.attn_queries.43",
"model.llm.attn_res.attn_queries.44",
"model.llm.attn_res.attn_queries.45",
"model.llm.attn_res.attn_queries.46",
"model.llm.attn_res.attn_queries.5",
"model.llm.attn_res.attn_queries.6",
"model.llm.attn_res.attn_queries.7",
"model.llm.attn_res.attn_queries.8",
"model.llm.attn_res.attn_queries.9",
"model.llm.attn_res.mlp_queries.0",
"model.llm.attn_res.mlp_queries.1",
"model.llm.attn_res.mlp_queries.10",
"model.llm.attn_res.mlp_queries.11",
"model.llm.attn_res.mlp_queries.12",
"model.llm.attn_res.mlp_queries.13",
"model.llm.attn_res.mlp_queries.14",
"model.llm.attn_res.mlp_queries.15",
"model.llm.attn_res.mlp_queries.16",
"model.llm.attn_res.mlp_queries.17",
"model.llm.attn_res.mlp_queries.18",
"model.llm.attn_res.mlp_queries.19",
"model.llm.attn_res.mlp_queries.2",
"model.llm.attn_res.mlp_queries.20",
"model.llm.attn_res.mlp_queries.21",
"model.llm.attn_res.mlp_queries.22",
"model.llm.attn_res.mlp_queries.23",
"model.llm.attn_res.mlp_queries.24",
"model.llm.attn_res.mlp_queries.25",
"model.llm.attn_res.mlp_queries.26",
"model.llm.attn_res.mlp_queries.27",
"model.llm.attn_res.mlp_queries.28",
"model.llm.attn_res.mlp_queries.29",
"model.llm.attn_res.mlp_queries.3",
"model.llm.attn_res.mlp_queries.30",
"model.llm.attn_res.mlp_queries.31",
"model.llm.attn_res.mlp_queries.32",
"model.llm.attn_res.mlp_queries.33",
"model.llm.attn_res.mlp_queries.34",
"model.llm.attn_res.mlp_queries.35",
"model.llm.attn_res.mlp_queries.36",
"model.llm.attn_res.mlp_queries.37",
"model.llm.attn_res.mlp_queries.38",
"model.llm.attn_res.mlp_queries.39",
"model.llm.attn_res.mlp_queries.4",
"model.llm.attn_res.mlp_queries.40",
"model.llm.attn_res.mlp_queries.41",
"model.llm.attn_res.mlp_queries.42",
"model.llm.attn_res.mlp_queries.43",
"model.llm.attn_res.mlp_queries.44",
"model.llm.attn_res.mlp_queries.45",
"model.llm.attn_res.mlp_queries.46",
"model.llm.attn_res.mlp_queries.47",
"model.llm.attn_res.mlp_queries.5",
"model.llm.attn_res.mlp_queries.6",
"model.llm.attn_res.mlp_queries.7",
"model.llm.attn_res.mlp_queries.8",
"model.llm.attn_res.mlp_queries.9",
"model.llm.attn_res.output_query",
"model.llm.layers.0.mlp.gate.bias",
"model.llm.layers.1.mlp.gate.bias",
"model.llm.layers.10.mlp.gate.bias",
"model.llm.layers.11.mlp.gate.bias",
"model.llm.layers.12.mlp.gate.bias",
"model.llm.layers.13.mlp.gate.bias",
"model.llm.layers.14.mlp.gate.bias",
"model.llm.layers.15.mlp.gate.bias",
"model.llm.layers.16.mlp.gate.bias",
"model.llm.layers.17.mlp.gate.bias",
"model.llm.layers.18.mlp.gate.bias",
"model.llm.layers.19.mlp.gate.bias",
"model.llm.layers.2.mlp.gate.bias",
"model.llm.layers.20.mlp.gate.bias",
"model.llm.layers.21.mlp.gate.bias",
"model.llm.layers.22.mlp.gate.bias",
"model.llm.layers.23.mlp.gate.bias",
"model.llm.layers.24.mlp.gate.bias",
"model.llm.layers.25.mlp.gate.bias",
"model.llm.layers.26.mlp.gate.bias",
"model.llm.layers.27.mlp.gate.bias",
"model.llm.layers.28.mlp.gate.bias",
"model.llm.layers.29.mlp.gate.bias",
"model.llm.layers.3.mlp.gate.bias",
"model.llm.layers.30.mlp.gate.bias",
"model.llm.layers.31.mlp.gate.bias",
"model.llm.layers.32.mlp.gate.bias",
"model.llm.layers.33.mlp.gate.bias",
"model.llm.layers.34.mlp.gate.bias",
"model.llm.layers.35.mlp.gate.bias",
"model.llm.layers.36.mlp.gate.bias",
"model.llm.layers.37.mlp.gate.bias",
"model.llm.layers.38.mlp.gate.bias",
"model.llm.layers.39.mlp.gate.bias",
"model.llm.layers.4.mlp.gate.bias",
"model.llm.layers.40.mlp.gate.bias",
"model.llm.layers.41.mlp.gate.bias",
"model.llm.layers.42.mlp.gate.bias",
"model.llm.layers.43.mlp.gate.bias",
"model.llm.layers.44.mlp.gate.bias",
"model.llm.layers.45.mlp.gate.bias",
"model.llm.layers.46.mlp.gate.bias",
"model.llm.layers.47.mlp.gate.bias",
"model.llm.layers.5.mlp.gate.bias",
"model.llm.layers.6.mlp.gate.bias",
"model.llm.layers.7.mlp.gate.bias",
"model.llm.layers.8.mlp.gate.bias",
"model.llm.layers.9.mlp.gate.bias"
],
"weight_block_shape": [],
"weight_transform": "cast"
},
{
"compute_dtype": "bfloat16",
"output_dtype": "bfloat16",
"required_activation_quantization": null,
"tensors": [
"model.llm.embed.weight",
"model.llm.embed_norm.weight",
"model.llm.layers.0.attn.attn_sink_proj.weight",
"model.llm.layers.0.attn.gate_attn.weight",
"model.llm.layers.0.attn.k_sconv.weight",
"model.llm.layers.0.attn.q_l2_clip.scale",
"model.llm.layers.0.attn.rel_l2_clip.scale",
"model.llm.layers.0.attn.rel_logits_proj.proj",
"model.llm.layers.0.attn.v_sconv.weight",
"model.llm.layers.0.attn.wk_dv.weight",
"model.llm.layers.0.attn.wo_ud.weight",
"model.llm.layers.0.attn.wq_du.weight",
"model.llm.layers.0.attn.wr_du.weight",
"model.llm.layers.0.attn.wv_dv.weight",
"model.llm.layers.0.attn_norm.weight",
"model.llm.layers.0.attn_sconv.weight",
"model.llm.layers.0.mlp.gate.weight",
"model.llm.layers.0.mlp_norm.weight",
"model.llm.layers.0.mlp_sconv.weight",
"model.llm.layers.1.attn.attn_sink_proj.weight",
"model.llm.layers.1.attn.gate_attn.weight",
"model.llm.layers.1.attn.k_sconv.weight",
"model.llm.layers.1.attn.q_l2_clip.scale",
"model.llm.layers.1.attn.rel_l2_clip.scale",
"model.llm.layers.1.attn.rel_logits_proj.proj",
"model.llm.layers.1.attn.v_sconv.weight",
"model.llm.layers.1.attn.wk_dv.weight",
"model.llm.layers.1.attn.wo_ud.weight",
"model.llm.layers.1.attn.wq_du.weight",
"model.llm.layers.1.attn.wr_du.weight",
"model.llm.layers.1.attn.wv_dv.weight",
"model.llm.layers.1.attn_norm.weight",
"model.llm.layers.1.attn_sconv.weight",
"model.llm.layers.1.mlp.gate.weight",
"model.llm.layers.1.mlp_norm.weight",
"model.llm.layers.1.mlp_sconv.weight",
"model.llm.layers.10.attn.attn_sink_proj.weight",
"model.llm.layers.10.attn.gate_attn.weight",
"model.llm.layers.10.attn.k_sconv.weight",
"model.llm.layers.10.attn.q_l2_clip.scale",
"model.llm.layers.10.attn.rel_l2_clip.scale",
"model.llm.layers.10.attn.rel_logits_proj.proj",
"model.llm.layers.10.attn.v_sconv.weight",
"model.llm.layers.10.attn.wk_dv.weight",
"model.llm.layers.10.attn.wo_ud.weight",
"model.llm.layers.10.attn.wq_du.weight",
"model.llm.layers.10.attn.wr_du.weight",
"model.llm.layers.10.attn.wv_dv.weight",
"model.llm.layers.10.attn_norm.weight",
"model.llm.layers.10.attn_sconv.weight",
"model.llm.layers.10.mlp.gate.weight",
"model.llm.layers.10.mlp_norm.weight",
"model.llm.layers.10.mlp_sconv.weight",
"model.llm.layers.11.attn.attn_sink_proj.weight",
"model.llm.layers.11.attn.gate_attn.weight",
"model.llm.layers.11.attn.k_sconv.static_w",
"model.llm.layers.11.attn.k_sconv.weight_proj",
"model.llm.layers.11.attn.q_l2_clip.scale",
"model.llm.layers.11.attn.rel_l2_clip.scale",
"model.llm.layers.11.attn.rel_logits_proj.proj",
"model.llm.layers.11.attn.v_sconv.static_w",
"model.llm.layers.11.attn.v_sconv.weight_proj",
"model.llm.layers.11.attn.wk_dv.weight",
"model.llm.layers.11.attn.wo_ud.weight",
"model.llm.layers.11.attn.wq_du.weight",
"model.llm.layers.11.attn.wr_du.weight",
"model.llm.layers.11.attn.wv_dv.weight",
"model.llm.layers.11.attn_sconv.weight",
"model.llm.layers.11.k_input_norm.weight",
"model.llm.layers.11.mlp.gate.weight",
"model.llm.layers.11.mlp_norm.weight",
"model.llm.layers.11.mlp_sconv.weight",
"model.llm.layers.11.q_input_norm.weight",
"model.llm.layers.11.v_input_norm.weight",
"model.llm.layers.12.attn.attn_sink_proj.weight",
"model.llm.layers.12.attn.gate_attn.weight",
"model.llm.layers.12.attn.k_sconv.weight",
"model.llm.layers.12.attn.q_l2_clip.scale",
"model.llm.layers.12.attn.rel_l2_clip.scale",
"model.llm.layers.12.attn.rel_logits_proj.proj",
"model.llm.layers.12.attn.v_sconv.weight",
"model.llm.layers.12.attn.wk_dv.weight",
"model.llm.layers.12.attn.wo_ud.weight",
"model.llm.layers.12.attn.wq_du.weight",
"model.llm.layers.12.attn.wr_du.weight",
"model.llm.layers.12.attn.wv_dv.weight",
"model.llm.layers.12.attn_norm.weight",
"model.llm.layers.12.attn_sconv.weight",
"model.llm.layers.12.mlp.gate.weight",
"model.llm.layers.12.mlp_norm.weight",
"model.llm.layers.12.mlp_sconv.weight",
"model.llm.layers.13.attn.attn_sink_proj.weight",
"model.llm.layers.13.attn.gate_attn.weight",
"model.llm.layers.13.attn.k_sconv.weight",
"model.llm.layers.13.attn.q_l2_clip.scale",
"model.llm.layers.13.attn.rel_l2_clip.scale",
"model.llm.layers.13.attn.rel_logits_proj.proj",
"model.llm.layers.13.attn.v_sconv.weight",
"model.llm.layers.13.attn.wk_dv.weight",
"model.llm.layers.13.attn.wo_ud.weight",
"model.llm.layers.13.attn.wq_du.weight",
"model.llm.layers.13.attn.wr_du.weight",
"model.llm.layers.13.attn.wv_dv.weight",
"model.llm.layers.13.attn_norm.weight",
"model.llm.layers.13.attn_sconv.weight",
"model.llm.layers.13.mlp.gate.weight",
"model.llm.layers.13.mlp_norm.weight",
"model.llm.layers.13.mlp_sconv.weight",
"model.llm.layers.14.attn.attn_sink_proj.weight",
"model.llm.layers.14.attn.gate_attn.weight",
"model.llm.layers.14.attn.k_sconv.weight",
"model.llm.layers.14.attn.q_l2_clip.scale",
"model.llm.layers.14.attn.rel_l2_clip.scale",
"model.llm.layers.14.attn.rel_logits_proj.proj",
"model.llm.layers.14.attn.v_sconv.weight",
"model.llm.layers.14.attn.wk_dv.weight",
"model.llm.layers.14.attn.wo_ud.weight",
"model.llm.layers.14.attn.wq_du.weight",
"model.llm.layers.14.attn.wr_du.weight",
"model.llm.layers.14.attn.wv_dv.weight",
"model.llm.layers.14.attn_norm.weight",
"model.llm.layers.14.attn_sconv.weight",
"model.llm.layers.14.mlp.gate.weight",
"model.llm.layers.14.mlp_norm.weight",
"model.llm.layers.14.mlp_sconv.weight",
"model.llm.layers.15.attn.attn_sink_proj.weight",
"model.llm.layers.15.attn.gate_attn.weight",
"model.llm.layers.15.attn.k_sconv.weight",
"model.llm.layers.15.attn.q_l2_clip.scale",
"model.llm.layers.15.attn.rel_l2_clip.scale",
"model.llm.layers.15.attn.rel_logits_proj.proj",
"model.llm.layers.15.attn.v_sconv.weight",
"model.llm.layers.15.attn.wk_dv.weight",
"model.llm.layers.15.attn.wo_ud.weight",
"model.llm.layers.15.attn.wq_du.weight",
"model.llm.layers.15.attn.wr_du.weight",
"model.llm.layers.15.attn.wv_dv.weight",
"model.llm.layers.15.attn_norm.weight",
"model.llm.layers.15.attn_sconv.weight",
"model.llm.layers.15.mlp.gate.weight",
"model.llm.layers.15.mlp_norm.weight",
"model.llm.layers.15.mlp_sconv.weight",
"model.llm.layers.16.attn.attn_sink_proj.weight",
"model.llm.layers.16.attn.gate_attn.weight",
"model.llm.layers.16.attn.k_sconv.weight",
"model.llm.layers.16.attn.q_l2_clip.scale",
"model.llm.layers.16.attn.rel_l2_clip.scale",
"model.llm.layers.16.attn.rel_logits_proj.proj",
"model.llm.layers.16.attn.v_sconv.weight",
"model.llm.layers.16.attn.wk_dv.weight",
"model.llm.layers.16.attn.wo_ud.weight",
"model.llm.layers.16.attn.wq_du.weight",
"model.llm.layers.16.attn.wr_du.weight",
"model.llm.layers.16.attn.wv_dv.weight",
"model.llm.layers.16.attn_norm.weight",
"model.llm.layers.16.attn_sconv.weight",
"model.llm.layers.16.mlp.gate.weight",
"model.llm.layers.16.mlp_norm.weight",
"model.llm.layers.16.mlp_sconv.weight",
"model.llm.layers.17.attn.attn_sink_proj.weight",
"model.llm.layers.17.attn.gate_attn.weight",
"model.llm.layers.17.attn.k_sconv.static_w",
"model.llm.layers.17.attn.k_sconv.weight_proj",
"model.llm.layers.17.attn.q_l2_clip.scale",
"model.llm.layers.17.attn.rel_l2_clip.scale",
"model.llm.layers.17.attn.rel_logits_proj.proj",
"model.llm.layers.17.attn.v_sconv.static_w",
"model.llm.layers.17.attn.v_sconv.weight_proj",
"model.llm.layers.17.attn.wk_dv.weight",
"model.llm.layers.17.attn.wo_ud.weight",
"model.llm.layers.17.attn.wq_du.weight",
"model.llm.layers.17.attn.wr_du.weight",
"model.llm.layers.17.attn.wv_dv.weight",
"model.llm.layers.17.attn_sconv.weight",
"model.llm.layers.17.k_input_norm.weight",
"model.llm.layers.17.mlp.gate.weight",
"model.llm.layers.17.mlp_norm.weight",
"model.llm.layers.17.mlp_sconv.weight",
"model.llm.layers.17.q_input_norm.weight",
"model.llm.layers.17.v_input_norm.weight",
"model.llm.layers.18.attn.attn_sink_proj.weight",
"model.llm.layers.18.attn.gate_attn.weight",
"model.llm.layers.18.attn.k_sconv.weight",
"model.llm.layers.18.attn.q_l2_clip.scale",
"model.llm.layers.18.attn.rel_l2_clip.scale",
"model.llm.layers.18.attn.rel_logits_proj.proj",
"model.llm.layers.18.attn.v_sconv.weight",
"model.llm.layers.18.attn.wk_dv.weight",
"model.llm.layers.18.attn.wo_ud.weight",
"model.llm.layers.18.attn.wq_du.weight",
"model.llm.layers.18.attn.wr_du.weight",
"model.llm.layers.18.attn.wv_dv.weight",
"model.llm.layers.18.attn_norm.weight",
"model.llm.layers.18.attn_sconv.weight",
"model.llm.layers.18.mlp.gate.weight",
"model.llm.layers.18.mlp_norm.weight",
"model.llm.layers.18.mlp_sconv.weight",
"model.llm.layers.19.attn.attn_sink_proj.weight",
"model.llm.layers.19.attn.gate_attn.weight",
"model.llm.layers.19.attn.k_sconv.weight",
"model.llm.layers.19.attn.q_l2_clip.scale",
"model.llm.layers.19.attn.rel_l2_clip.scale",
"model.llm.layers.19.attn.rel_logits_proj.proj",
"model.llm.layers.19.attn.v_sconv.weight",
"model.llm.layers.19.attn.wk_dv.weight",
"model.llm.layers.19.attn.wo_ud.weight",
"model.llm.layers.19.attn.wq_du.weight",
"model.llm.layers.19.attn.wr_du.weight",
"model.llm.layers.19.attn.wv_dv.weight",
"model.llm.layers.19.attn_norm.weight",
"model.llm.layers.19.attn_sconv.weight",
"model.llm.layers.19.mlp.gate.weight",
"model.llm.layers.19.mlp_norm.weight",
"model.llm.layers.19.mlp_sconv.weight",
"model.llm.layers.2.attn.attn_sink_proj.weight",
"model.llm.layers.2.attn.gate_attn.weight",
"model.llm.layers.2.attn.k_sconv.weight",
"model.llm.layers.2.attn.q_l2_clip.scale",
"model.llm.layers.2.attn.rel_l2_clip.scale",
"model.llm.layers.2.attn.rel_logits_proj.proj",
"model.llm.layers.2.attn.v_sconv.weight",
"model.llm.layers.2.attn.wk_dv.weight",
"model.llm.layers.2.attn.wo_ud.weight",
"model.llm.layers.2.attn.wq_du.weight",
"model.llm.layers.2.attn.wr_du.weight",
"model.llm.layers.2.attn.wv_dv.weight",
"model.llm.layers.2.attn_norm.weight",
"model.llm.layers.2.attn_sconv.weight",
"model.llm.layers.2.mlp.gate.weight",
"model.llm.layers.2.mlp_norm.weight",
"model.llm.layers.2.mlp_sconv.weight",
"model.llm.layers.20.attn.attn_sink_proj.weight",
"model.llm.layers.20.attn.gate_attn.weight",
"model.llm.layers.20.attn.k_sconv.weight",
"model.llm.layers.20.attn.q_l2_clip.scale",
"model.llm.layers.20.attn.rel_l2_clip.scale",
"model.llm.layers.20.attn.rel_logits_proj.proj",
"model.llm.layers.20.attn.v_sconv.weight",
"model.llm.layers.20.attn.wk_dv.weight",
"model.llm.layers.20.attn.wo_ud.weight",
"model.llm.layers.20.attn.wq_du.weight",
"model.llm.layers.20.attn.wr_du.weight",
"model.llm.layers.20.attn.wv_dv.weight",
"model.llm.layers.20.attn_norm.weight",
"model.llm.layers.20.attn_sconv.weight",
"model.llm.layers.20.mlp.gate.weight",
"model.llm.layers.20.mlp_norm.weight",
"model.llm.layers.20.mlp_sconv.weight",
"model.llm.layers.21.attn.attn_sink_proj.weight",
"model.llm.layers.21.attn.gate_attn.weight",
"model.llm.layers.21.attn.k_sconv.weight",
"model.llm.layers.21.attn.q_l2_clip.scale",
"model.llm.layers.21.attn.rel_l2_clip.scale",
"model.llm.layers.21.attn.rel_logits_proj.proj",
"model.llm.layers.21.attn.v_sconv.weight",
"model.llm.layers.21.attn.wk_dv.weight",
"model.llm.layers.21.attn.wo_ud.weight",
"model.llm.layers.21.attn.wq_du.weight",
"model.llm.layers.21.attn.wr_du.weight",
"model.llm.layers.21.attn.wv_dv.weight",
"model.llm.layers.21.attn_norm.weight",
"model.llm.layers.21.attn_sconv.weight",
"model.llm.layers.21.mlp.gate.weight",
"model.llm.layers.21.mlp_norm.weight",
"model.llm.layers.21.mlp_sconv.weight",
"model.llm.layers.22.attn.attn_sink_proj.weight",
"model.llm.layers.22.attn.gate_attn.weight",
"model.llm.layers.22.attn.k_sconv.weight",
"model.llm.layers.22.attn.q_l2_clip.scale",
"model.llm.layers.22.attn.rel_l2_clip.scale",
"model.llm.layers.22.attn.rel_logits_proj.proj",
"model.llm.layers.22.attn.v_sconv.weight",
"model.llm.layers.22.attn.wk_dv.weight",
"model.llm.layers.22.attn.wo_ud.weight",
"model.llm.layers.22.attn.wq_du.weight",
"model.llm.layers.22.attn.wr_du.weight",
"model.llm.layers.22.attn.wv_dv.weight",
"model.llm.layers.22.attn_norm.weight",
"model.llm.layers.22.attn_sconv.weight",
"model.llm.layers.22.mlp.gate.weight",
"model.llm.layers.22.mlp_norm.weight",
"model.llm.layers.22.mlp_sconv.weight",
"model.llm.layers.23.attn.attn_sink_proj.weight",
"model.llm.layers.23.attn.gate_attn.weight",
"model.llm.layers.23.attn.k_sconv.static_w",
"model.llm.layers.23.attn.k_sconv.weight_proj",
"model.llm.layers.23.attn.q_l2_clip.scale",
"model.llm.layers.23.attn.rel_l2_clip.scale",
"model.llm.layers.23.attn.rel_logits_proj.proj",
"model.llm.layers.23.attn.v_sconv.static_w",
"model.llm.layers.23.attn.v_sconv.weight_proj",
"model.llm.layers.23.attn.wk_dv.weight",
"model.llm.layers.23.attn.wo_ud.weight",
"model.llm.layers.23.attn.wq_du.weight",
"model.llm.layers.23.attn.wr_du.weight",
"model.llm.layers.23.attn.wv_dv.weight",
"model.llm.layers.23.attn_sconv.weight",
"model.llm.layers.23.k_input_norm.weight",
"model.llm.layers.23.mlp.gate.weight",
"model.llm.layers.23.mlp_norm.weight",
"model.llm.layers.23.mlp_sconv.weight",
"model.llm.layers.23.q_input_norm.weight",
"model.llm.layers.23.v_input_norm.weight",
"model.llm.layers.24.attn.attn_sink_proj.weight",
"model.llm.layers.24.attn.gate_attn.weight",
"model.llm.layers.24.attn.k_sconv.weight",
"model.llm.layers.24.attn.q_l2_clip.scale",
"model.llm.layers.24.attn.rel_l2_clip.scale",
"model.llm.layers.24.attn.rel_logits_proj.proj",
"model.llm.layers.24.attn.v_sconv.weight",
"model.llm.layers.24.attn.wk_dv.weight",
"model.llm.layers.24.attn.wo_ud.weight",
"model.llm.layers.24.attn.wq_du.weight",
"model.llm.layers.24.attn.wr_du.weight",
"model.llm.layers.24.attn.wv_dv.weight",
"model.llm.layers.24.attn_norm.weight",
"model.llm.layers.24.attn_sconv.weight",
"model.llm.layers.24.mlp.gate.weight",
"model.llm.layers.24.mlp_norm.weight",
"model.llm.layers.24.mlp_sconv.weight",
"model.llm.layers.25.attn.attn_sink_proj.weight",
"model.llm.layers.25.attn.gate_attn.weight",
"model.llm.layers.25.attn.k_sconv.weight",
"model.llm.layers.25.attn.q_l2_clip.scale",
"model.llm.layers.25.attn.rel_l2_clip.scale",
"model.llm.layers.25.attn.rel_logits_proj.proj",
"model.llm.layers.25.attn.v_sconv.weight",
"model.llm.layers.25.attn.wk_dv.weight",
"model.llm.layers.25.attn.wo_ud.weight",
"model.llm.layers.25.attn.wq_du.weight",
"model.llm.layers.25.attn.wr_du.weight",
"model.llm.layers.25.attn.wv_dv.weight",
"model.llm.layers.25.attn_norm.weight",
"model.llm.layers.25.attn_sconv.weight",
"model.llm.layers.25.mlp.gate.weight",
"model.llm.layers.25.mlp_norm.weight",
"model.llm.layers.25.mlp_sconv.weight",
"model.llm.layers.26.attn.attn_sink_proj.weight",
"model.llm.layers.26.attn.gate_attn.weight",
"model.llm.layers.26.attn.k_sconv.weight",
"model.llm.layers.26.attn.q_l2_clip.scale",
"model.llm.layers.26.attn.rel_l2_clip.scale",
"model.llm.layers.26.attn.rel_logits_proj.proj",
"model.llm.layers.26.attn.v_sconv.weight",
"model.llm.layers.26.attn.wk_dv.weight",
"model.llm.layers.26.attn.wo_ud.weight",
"model.llm.layers.26.attn.wq_du.weight",
"model.llm.layers.26.attn.wr_du.weight",
"model.llm.layers.26.attn.wv_dv.weight",
"model.llm.layers.26.attn_norm.weight",
"model.llm.layers.26.attn_sconv.weight",
"model.llm.layers.26.mlp.gate.weight",
"model.llm.layers.26.mlp_norm.weight",
"model.llm.layers.26.mlp_sconv.weight",
"model.llm.layers.27.attn.attn_sink_proj.weight",
"model.llm.layers.27.attn.gate_attn.weight",
"model.llm.layers.27.attn.k_sconv.weight",
"model.llm.layers.27.attn.q_l2_clip.scale",
"model.llm.layers.27.attn.rel_l2_clip.scale",
"model.llm.layers.27.attn.rel_logits_proj.proj",
"model.llm.layers.27.attn.v_sconv.weight",
"model.llm.layers.27.attn.wk_dv.weight",
"model.llm.layers.27.attn.wo_ud.weight",
"model.llm.layers.27.attn.wq_du.weight",
"model.llm.layers.27.attn.wr_du.weight",
"model.llm.layers.27.attn.wv_dv.weight",
"model.llm.layers.27.attn_norm.weight",
"model.llm.layers.27.attn_sconv.weight",
"model.llm.layers.27.mlp.gate.weight",
"model.llm.layers.27.mlp_norm.weight",
"model.llm.layers.27.mlp_sconv.weight",
"model.llm.layers.28.attn.attn_sink_proj.weight",
"model.llm.layers.28.attn.gate_attn.weight",
"model.llm.layers.28.attn.k_sconv.weight",
"model.llm.layers.28.attn.q_l2_clip.scale",
"model.llm.layers.28.attn.rel_l2_clip.scale",
"model.llm.layers.28.attn.rel_logits_proj.proj",
"model.llm.layers.28.attn.v_sconv.weight",
"model.llm.layers.28.attn.wk_dv.weight",
"model.llm.layers.28.attn.wo_ud.weight",
"model.llm.layers.28.attn.wq_du.weight",
"model.llm.layers.28.attn.wr_du.weight",
"model.llm.layers.28.attn.wv_dv.weight",
"model.llm.layers.28.attn_norm.weight",
"model.llm.layers.28.attn_sconv.weight",
"model.llm.layers.28.mlp.gate.weight",
"model.llm.layers.28.mlp_norm.weight",
"model.llm.layers.28.mlp_sconv.weight",
"model.llm.layers.29.attn.attn_sink_proj.weight",
"model.llm.layers.29.attn.gate_attn.weight",
"model.llm.layers.29.attn.k_sconv.static_w",
"model.llm.layers.29.attn.k_sconv.weight_proj",
"model.llm.layers.29.attn.q_l2_clip.scale",
"model.llm.layers.29.attn.rel_l2_clip.scale",
"model.llm.layers.29.attn.rel_logits_proj.proj",
"model.llm.layers.29.attn.v_sconv.static_w",
"model.llm.layers.29.attn.v_sconv.weight_proj",
"model.llm.layers.29.attn.wk_dv.weight",
"model.llm.layers.29.attn.wo_ud.weight",
"model.llm.layers.29.attn.wq_du.weight",
"model.llm.layers.29.attn.wr_du.weight",
"model.llm.layers.29.attn.wv_dv.weight",
"model.llm.layers.29.attn_sconv.weight",
"model.llm.layers.29.k_input_norm.weight",
"model.llm.layers.29.mlp.gate.weight",
"model.llm.layers.29.mlp_norm.weight",
"model.llm.layers.29.mlp_sconv.weight",
"model.llm.layers.29.q_input_norm.weight",
"model.llm.layers.29.v_input_norm.weight",
"model.llm.layers.3.attn.attn_sink_proj.weight",
"model.llm.layers.3.attn.gate_attn.weight",
"model.llm.layers.3.attn.k_sconv.weight",
"model.llm.layers.3.attn.q_l2_clip.scale",
"model.llm.layers.3.attn.rel_l2_clip.scale",
"model.llm.layers.3.attn.rel_logits_proj.proj",
"model.llm.layers.3.attn.v_sconv.weight",
"model.llm.layers.3.attn.wk_dv.weight",
"model.llm.layers.3.attn.wo_ud.weight",
"model.llm.layers.3.attn.wq_du.weight",
"model.llm.layers.3.attn.wr_du.weight",
"model.llm.layers.3.attn.wv_dv.weight",
"model.llm.layers.3.attn_norm.weight",
"model.llm.layers.3.attn_sconv.weight",
"model.llm.layers.3.mlp.gate.weight",
"model.llm.layers.3.mlp_norm.weight",
"model.llm.layers.3.mlp_sconv.weight",
"model.llm.layers.30.attn.attn_sink_proj.weight",
"model.llm.layers.30.attn.gate_attn.weight",
"model.llm.layers.30.attn.k_sconv.weight",
"model.llm.layers.30.attn.q_l2_clip.scale",
"model.llm.layers.30.attn.rel_l2_clip.scale",
"model.llm.layers.30.attn.rel_logits_proj.proj",
"model.llm.layers.30.attn.v_sconv.weight",
"model.llm.layers.30.attn.wk_dv.weight",
"model.llm.layers.30.attn.wo_ud.weight",
"model.llm.layers.30.attn.wq_du.weight",
"model.llm.layers.30.attn.wr_du.weight",
"model.llm.layers.30.attn.wv_dv.weight",
"model.llm.layers.30.attn_norm.weight",
"model.llm.layers.30.attn_sconv.weight",
"model.llm.layers.30.mlp.gate.weight",
"model.llm.layers.30.mlp_norm.weight",
"model.llm.layers.30.mlp_sconv.weight",
"model.llm.layers.31.attn.attn_sink_proj.weight",
"model.llm.layers.31.attn.gate_attn.weight",
"model.llm.layers.31.attn.k_sconv.weight",
"model.llm.layers.31.attn.q_l2_clip.scale",
"model.llm.layers.31.attn.rel_l2_clip.scale",
"model.llm.layers.31.attn.rel_logits_proj.proj",
"model.llm.layers.31.attn.v_sconv.weight",
"model.llm.layers.31.attn.wk_dv.weight",
"model.llm.layers.31.attn.wo_ud.weight",
"model.llm.layers.31.attn.wq_du.weight",
"model.llm.layers.31.attn.wr_du.weight",
"model.llm.layers.31.attn.wv_dv.weight",
"model.llm.layers.31.attn_norm.weight",
"model.llm.layers.31.attn_sconv.weight",
"model.llm.layers.31.mlp.gate.weight",
"model.llm.layers.31.mlp_norm.weight",
"model.llm.layers.31.mlp_sconv.weight",
"model.llm.layers.32.attn.attn_sink_proj.weight",
"model.llm.layers.32.attn.gate_attn.weight",
"model.llm.layers.32.attn.k_sconv.weight",
"model.llm.layers.32.attn.q_l2_clip.scale",
"model.llm.layers.32.attn.rel_l2_clip.scale",
"model.llm.layers.32.attn.rel_logits_proj.proj",
"model.llm.layers.32.attn.v_sconv.weight",
"model.llm.layers.32.attn.wk_dv.weight",
"model.llm.layers.32.attn.wo_ud.weight",
"model.llm.layers.32.attn.wq_du.weight",
"model.llm.layers.32.attn.wr_du.weight",
"model.llm.layers.32.attn.wv_dv.weight",
"model.llm.layers.32.attn_norm.weight",
"model.llm.layers.32.attn_sconv.weight",
"model.llm.layers.32.mlp.gate.weight",
"model.llm.layers.32.mlp_norm.weight",
"model.llm.layers.32.mlp_sconv.weight",
"model.llm.layers.33.attn.attn_sink_proj.weight",
"model.llm.layers.33.attn.gate_attn.weight",
"model.llm.layers.33.attn.k_sconv.weight",
"model.llm.layers.33.attn.q_l2_clip.scale",
"model.llm.layers.33.attn.rel_l2_clip.scale",
"model.llm.layers.33.attn.rel_logits_proj.proj",
"model.llm.layers.33.attn.v_sconv.weight",
"model.llm.layers.33.attn.wk_dv.weight",
"model.llm.layers.33.attn.wo_ud.weight",
"model.llm.layers.33.attn.wq_du.weight",
"model.llm.layers.33.attn.wr_du.weight",
"model.llm.layers.33.attn.wv_dv.weight",
"model.llm.layers.33.attn_norm.weight",
"model.llm.layers.33.attn_sconv.weight",
"model.llm.layers.33.mlp.gate.weight",
"model.llm.layers.33.mlp_norm.weight",
"model.llm.layers.33.mlp_sconv.weight",
"model.llm.layers.34.attn.attn_sink_proj.weight",
"model.llm.layers.34.attn.gate_attn.weight",
"model.llm.layers.34.attn.k_sconv.weight",
"model.llm.layers.34.attn.q_l2_clip.scale",
"model.llm.layers.34.attn.rel_l2_clip.scale",
"model.llm.layers.34.attn.rel_logits_proj.proj",
"model.llm.layers.34.attn.v_sconv.weight",
"model.llm.layers.34.attn.wk_dv.weight",
"model.llm.layers.34.attn.wo_ud.weight",
"model.llm.layers.34.attn.wq_du.weight",
"model.llm.layers.34.attn.wr_du.weight",
"model.llm.layers.34.attn.wv_dv.weight",
"model.llm.layers.34.attn_norm.weight",
"model.llm.layers.34.attn_sconv.weight",
"model.llm.layers.34.mlp.gate.weight",
"model.llm.layers.34.mlp_norm.weight",
"model.llm.layers.34.mlp_sconv.weight",
"model.llm.layers.35.attn.attn_sink_proj.weight",
"model.llm.layers.35.attn.gate_attn.weight",
"model.llm.layers.35.attn.k_sconv.static_w",
"model.llm.layers.35.attn.k_sconv.weight_proj",
"model.llm.layers.35.attn.q_l2_clip.scale",
"model.llm.layers.35.attn.rel_l2_clip.scale",
"model.llm.layers.35.attn.rel_logits_proj.proj",
"model.llm.layers.35.attn.v_sconv.static_w",
"model.llm.layers.35.attn.v_sconv.weight_proj",
"model.llm.layers.35.attn.wk_dv.weight",
"model.llm.layers.35.attn.wo_ud.weight",
"model.llm.layers.35.attn.wq_du.weight",
"model.llm.layers.35.attn.wr_du.weight",
"model.llm.layers.35.attn.wv_dv.weight",
"model.llm.layers.35.attn_sconv.weight",
"model.llm.layers.35.k_input_norm.weight",
"model.llm.layers.35.mlp.gate.weight",
"model.llm.layers.35.mlp_norm.weight",
"model.llm.layers.35.mlp_sconv.weight",
"model.llm.layers.35.q_input_norm.weight",
"model.llm.layers.35.v_input_norm.weight",
"model.llm.layers.36.attn.attn_sink_proj.weight",
"model.llm.layers.36.attn.gate_attn.weight",
"model.llm.layers.36.attn.k_sconv.weight",
"model.llm.layers.36.attn.q_l2_clip.scale",
"model.llm.layers.36.attn.rel_l2_clip.scale",
"model.llm.layers.36.attn.rel_logits_proj.proj",
"model.llm.layers.36.attn.v_sconv.weight",
"model.llm.layers.36.attn.wk_dv.weight",
"model.llm.layers.36.attn.wo_ud.weight",
"model.llm.layers.36.attn.wq_du.weight",
"model.llm.layers.36.attn.wr_du.weight",
"model.llm.layers.36.attn.wv_dv.weight",
"model.llm.layers.36.attn_norm.weight",
"model.llm.layers.36.attn_sconv.weight",
"model.llm.layers.36.mlp.gate.weight",
"model.llm.layers.36.mlp_norm.weight",
"model.llm.layers.36.mlp_sconv.weight",
"model.llm.layers.37.attn.attn_sink_proj.weight",
"model.llm.layers.37.attn.gate_attn.weight",
"model.llm.layers.37.attn.k_sconv.weight",
"model.llm.layers.37.attn.q_l2_clip.scale",
"model.llm.layers.37.attn.rel_l2_clip.scale",
"model.llm.layers.37.attn.rel_logits_proj.proj",
"model.llm.layers.37.attn.v_sconv.weight",
"model.llm.layers.37.attn.wk_dv.weight",
"model.llm.layers.37.attn.wo_ud.weight",
"model.llm.layers.37.attn.wq_du.weight",
"model.llm.layers.37.attn.wr_du.weight",
"model.llm.layers.37.attn.wv_dv.weight",
"model.llm.layers.37.attn_norm.weight",
"model.llm.layers.37.attn_sconv.weight",
"model.llm.layers.37.mlp.gate.weight",
"model.llm.layers.37.mlp_norm.weight",
"model.llm.layers.37.mlp_sconv.weight",
"model.llm.layers.38.attn.attn_sink_proj.weight",
"model.llm.layers.38.attn.gate_attn.weight",
"model.llm.layers.38.attn.k_sconv.weight",
"model.llm.layers.38.attn.q_l2_clip.scale",
"model.llm.layers.38.attn.rel_l2_clip.scale",
"model.llm.layers.38.attn.rel_logits_proj.proj",
"model.llm.layers.38.attn.v_sconv.weight",
"model.llm.layers.38.attn.wk_dv.weight",
"model.llm.layers.38.attn.wo_ud.weight",
"model.llm.layers.38.attn.wq_du.weight",
"model.llm.layers.38.attn.wr_du.weight",
"model.llm.layers.38.attn.wv_dv.weight",
"model.llm.layers.38.attn_norm.weight",
"model.llm.layers.38.attn_sconv.weight",
"model.llm.layers.38.mlp.gate.weight",
"model.llm.layers.38.mlp_norm.weight",
"model.llm.layers.38.mlp_sconv.weight",
"model.llm.layers.39.attn.attn_sink_proj.weight",
"model.llm.layers.39.attn.gate_attn.weight",
"model.llm.layers.39.attn.k_sconv.weight",
"model.llm.layers.39.attn.q_l2_clip.scale",
"model.llm.layers.39.attn.rel_l2_clip.scale",
"model.llm.layers.39.attn.rel_logits_proj.proj",
"model.llm.layers.39.attn.v_sconv.weight",
"model.llm.layers.39.attn.wk_dv.weight",
"model.llm.layers.39.attn.wo_ud.weight",
"model.llm.layers.39.attn.wq_du.weight",
"model.llm.layers.39.attn.wr_du.weight",
"model.llm.layers.39.attn.wv_dv.weight",
"model.llm.layers.39.attn_norm.weight",
"model.llm.layers.39.attn_sconv.weight",
"model.llm.layers.39.mlp.gate.weight",
"model.llm.layers.39.mlp_norm.weight",
"model.llm.layers.39.mlp_sconv.weight",
"model.llm.layers.4.attn.attn_sink_proj.weight",
"model.llm.layers.4.attn.gate_attn.weight",
"model.llm.layers.4.attn.k_sconv.weight",
"model.llm.layers.4.attn.q_l2_clip.scale",
"model.llm.layers.4.attn.rel_l2_clip.scale",
"model.llm.layers.4.attn.rel_logits_proj.proj",
"model.llm.layers.4.attn.v_sconv.weight",
"model.llm.layers.4.attn.wk_dv.weight",
"model.llm.layers.4.attn.wo_ud.weight",
"model.llm.layers.4.attn.wq_du.weight",
"model.llm.layers.4.attn.wr_du.weight",
"model.llm.layers.4.attn.wv_dv.weight",
"model.llm.layers.4.attn_norm.weight",
"model.llm.layers.4.attn_sconv.weight",
"model.llm.layers.4.mlp.gate.weight",
"model.llm.layers.4.mlp_norm.weight",
"model.llm.layers.4.mlp_sconv.weight",
"model.llm.layers.40.attn.attn_sink_proj.weight",
"model.llm.layers.40.attn.gate_attn.weight",
"model.llm.layers.40.attn.k_sconv.weight",
"model.llm.layers.40.attn.q_l2_clip.scale",
"model.llm.layers.40.attn.rel_l2_clip.scale",
"model.llm.layers.40.attn.rel_logits_proj.proj",
"model.llm.layers.40.attn.v_sconv.weight",
"model.llm.layers.40.attn.wk_dv.weight",
"model.llm.layers.40.attn.wo_ud.weight",
"model.llm.layers.40.attn.wq_du.weight",
"model.llm.layers.40.attn.wr_du.weight",
"model.llm.layers.40.attn.wv_dv.weight",
"model.llm.layers.40.attn_norm.weight",
"model.llm.layers.40.attn_sconv.weight",
"model.llm.layers.40.mlp.gate.weight",
"model.llm.layers.40.mlp_norm.weight",
"model.llm.layers.40.mlp_sconv.weight",
"model.llm.layers.41.attn.attn_sink_proj.weight",
"model.llm.layers.41.attn.gate_attn.weight",
"model.llm.layers.41.attn.k_sconv.static_w",
"model.llm.layers.41.attn.k_sconv.weight_proj",
"model.llm.layers.41.attn.q_l2_clip.scale",
"model.llm.layers.41.attn.rel_l2_clip.scale",
"model.llm.layers.41.attn.rel_logits_proj.proj",
"model.llm.layers.41.attn.v_sconv.static_w",
"model.llm.layers.41.attn.v_sconv.weight_proj",
"model.llm.layers.41.attn.wk_dv.weight",
"model.llm.layers.41.attn.wo_ud.weight",
"model.llm.layers.41.attn.wq_du.weight",
"model.llm.layers.41.attn.wr_du.weight",
"model.llm.layers.41.attn.wv_dv.weight",
"model.llm.layers.41.attn_sconv.weight",
"model.llm.layers.41.k_input_norm.weight",
"model.llm.layers.41.mlp.gate.weight",
"model.llm.layers.41.mlp_norm.weight",
"model.llm.layers.41.mlp_sconv.weight",
"model.llm.layers.41.q_input_norm.weight",
"model.llm.layers.41.v_input_norm.weight",
"model.llm.layers.42.attn.attn_sink_proj.weight",
"model.llm.layers.42.attn.gate_attn.weight",
"model.llm.layers.42.attn.k_sconv.weight",
"model.llm.layers.42.attn.q_l2_clip.scale",
"model.llm.layers.42.attn.rel_l2_clip.scale",
"model.llm.layers.42.attn.rel_logits_proj.proj",
"model.llm.layers.42.attn.v_sconv.weight",
"model.llm.layers.42.attn.wk_dv.weight",
"model.llm.layers.42.attn.wo_ud.weight",
"model.llm.layers.42.attn.wq_du.weight",
"model.llm.layers.42.attn.wr_du.weight",
"model.llm.layers.42.attn.wv_dv.weight",
"model.llm.layers.42.attn_norm.weight",
"model.llm.layers.42.attn_sconv.weight",
"model.llm.layers.42.mlp.gate.weight",
"model.llm.layers.42.mlp_norm.weight",
"model.llm.layers.42.mlp_sconv.weight",
"model.llm.layers.43.attn.attn_sink_proj.weight",
"model.llm.layers.43.attn.gate_attn.weight",
"model.llm.layers.43.attn.k_sconv.weight",
"model.llm.layers.43.attn.q_l2_clip.scale",
"model.llm.layers.43.attn.rel_l2_clip.scale",
"model.llm.layers.43.attn.rel_logits_proj.proj",
"model.llm.layers.43.attn.v_sconv.weight",
"model.llm.layers.43.attn.wk_dv.weight",
"model.llm.layers.43.attn.wo_ud.weight",
"model.llm.layers.43.attn.wq_du.weight",
"model.llm.layers.43.attn.wr_du.weight",
"model.llm.layers.43.attn.wv_dv.weight",
"model.llm.layers.43.attn_norm.weight",
"model.llm.layers.43.attn_sconv.weight",
"model.llm.layers.43.mlp.gate.weight",
"model.llm.layers.43.mlp_norm.weight",
"model.llm.layers.43.mlp_sconv.weight",
"model.llm.layers.44.attn.attn_sink_proj.weight",
"model.llm.layers.44.attn.gate_attn.weight",
"model.llm.layers.44.attn.k_sconv.weight",
"model.llm.layers.44.attn.q_l2_clip.scale",
"model.llm.layers.44.attn.rel_l2_clip.scale",
"model.llm.layers.44.attn.rel_logits_proj.proj",
"model.llm.layers.44.attn.v_sconv.weight",
"model.llm.layers.44.attn.wk_dv.weight",
"model.llm.layers.44.attn.wo_ud.weight",
"model.llm.layers.44.attn.wq_du.weight",
"model.llm.layers.44.attn.wr_du.weight",
"model.llm.layers.44.attn.wv_dv.weight",
"model.llm.layers.44.attn_norm.weight",
"model.llm.layers.44.attn_sconv.weight",
"model.llm.layers.44.mlp.gate.weight",
"model.llm.layers.44.mlp_norm.weight",
"model.llm.layers.44.mlp_sconv.weight",
"model.llm.layers.45.attn.attn_sink_proj.weight",
"model.llm.layers.45.attn.gate_attn.weight",
"model.llm.layers.45.attn.k_sconv.weight",
"model.llm.layers.45.attn.q_l2_clip.scale",
"model.llm.layers.45.attn.rel_l2_clip.scale",
"model.llm.layers.45.attn.rel_logits_proj.proj",
"model.llm.layers.45.attn.v_sconv.weight",
"model.llm.layers.45.attn.wk_dv.weight",
"model.llm.layers.45.attn.wo_ud.weight",
"model.llm.layers.45.attn.wq_du.weight",
"model.llm.layers.45.attn.wr_du.weight",
"model.llm.layers.45.attn.wv_dv.weight",
"model.llm.layers.45.attn_norm.weight",
"model.llm.layers.45.attn_sconv.weight",
"model.llm.layers.45.mlp.gate.weight",
"model.llm.layers.45.mlp_norm.weight",
"model.llm.layers.45.mlp_sconv.weight",
"model.llm.layers.46.attn.attn_sink_proj.weight",
"model.llm.layers.46.attn.gate_attn.weight",
"model.llm.layers.46.attn.k_sconv.weight",
"model.llm.layers.46.attn.q_l2_clip.scale",
"model.llm.layers.46.attn.rel_l2_clip.scale",
"model.llm.layers.46.attn.rel_logits_proj.proj",
"model.llm.layers.46.attn.v_sconv.weight",
"model.llm.layers.46.attn.wk_dv.weight",
"model.llm.layers.46.attn.wo_ud.weight",
"model.llm.layers.46.attn.wq_du.weight",
"model.llm.layers.46.attn.wr_du.weight",
"model.llm.layers.46.attn.wv_dv.weight",
"model.llm.layers.46.attn_norm.weight",
"model.llm.layers.46.attn_sconv.weight",
"model.llm.layers.46.mlp.gate.weight",
"model.llm.layers.46.mlp_norm.weight",
"model.llm.layers.46.mlp_sconv.weight",
"model.llm.layers.47.attn.attn_sink_proj.weight",
"model.llm.layers.47.attn.gate_attn.weight",
"model.llm.layers.47.attn.k_sconv.static_w",
"model.llm.layers.47.attn.k_sconv.weight_proj",
"model.llm.layers.47.attn.q_l2_clip.scale",
"model.llm.layers.47.attn.rel_l2_clip.scale",
"model.llm.layers.47.attn.rel_logits_proj.proj",
"model.llm.layers.47.attn.v_sconv.static_w",
"model.llm.layers.47.attn.v_sconv.weight_proj",
"model.llm.layers.47.attn.wk_dv.weight",
"model.llm.layers.47.attn.wo_ud.weight",
"model.llm.layers.47.attn.wq_du.weight",
"model.llm.layers.47.attn.wr_du.weight",
"model.llm.layers.47.attn.wv_dv.weight",
"model.llm.layers.47.attn_sconv.weight",
"model.llm.layers.47.k_input_norm.weight",
"model.llm.layers.47.mlp.gate.weight",
"model.llm.layers.47.mlp_norm.weight",
"model.llm.layers.47.mlp_sconv.weight",
"model.llm.layers.47.q_input_norm.weight",
"model.llm.layers.47.v_input_norm.weight",
"model.llm.layers.5.attn.attn_sink_proj.weight",
"model.llm.layers.5.attn.gate_attn.weight",
"model.llm.layers.5.attn.k_sconv.static_w",
"model.llm.layers.5.attn.k_sconv.weight_proj",
"model.llm.layers.5.attn.q_l2_clip.scale",
"model.llm.layers.5.attn.rel_l2_clip.scale",
"model.llm.layers.5.attn.rel_logits_proj.proj",
"model.llm.layers.5.attn.v_sconv.static_w",
"model.llm.layers.5.attn.v_sconv.weight_proj",
"model.llm.layers.5.attn.wk_dv.weight",
"model.llm.layers.5.attn.wo_ud.weight",
"model.llm.layers.5.attn.wq_du.weight",
"model.llm.layers.5.attn.wr_du.weight",
"model.llm.layers.5.attn.wv_dv.weight",
"model.llm.layers.5.attn_sconv.weight",
"model.llm.layers.5.k_input_norm.weight",
"model.llm.layers.5.mlp.gate.weight",
"model.llm.layers.5.mlp_norm.weight",
"model.llm.layers.5.mlp_sconv.weight",
"model.llm.layers.5.q_input_norm.weight",
"model.llm.layers.5.v_input_norm.weight",
"model.llm.layers.6.attn.attn_sink_proj.weight",
"model.llm.layers.6.attn.gate_attn.weight",
"model.llm.layers.6.attn.k_sconv.weight",
"model.llm.layers.6.attn.q_l2_clip.scale",
"model.llm.layers.6.attn.rel_l2_clip.scale",
"model.llm.layers.6.attn.rel_logits_proj.proj",
"model.llm.layers.6.attn.v_sconv.weight",
"model.llm.layers.6.attn.wk_dv.weight",
"model.llm.layers.6.attn.wo_ud.weight",
"model.llm.layers.6.attn.wq_du.weight",
"model.llm.layers.6.attn.wr_du.weight",
"model.llm.layers.6.attn.wv_dv.weight",
"model.llm.layers.6.attn_norm.weight",
"model.llm.layers.6.attn_sconv.weight",
"model.llm.layers.6.mlp.gate.weight",
"model.llm.layers.6.mlp_norm.weight",
"model.llm.layers.6.mlp_sconv.weight",
"model.llm.layers.7.attn.attn_sink_proj.weight",
"model.llm.layers.7.attn.gate_attn.weight",
"model.llm.layers.7.attn.k_sconv.weight",
"model.llm.layers.7.attn.q_l2_clip.scale",
"model.llm.layers.7.attn.rel_l2_clip.scale",
"model.llm.layers.7.attn.rel_logits_proj.proj",
"model.llm.layers.7.attn.v_sconv.weight",
"model.llm.layers.7.attn.wk_dv.weight",
"model.llm.layers.7.attn.wo_ud.weight",
"model.llm.layers.7.attn.wq_du.weight",
"model.llm.layers.7.attn.wr_du.weight",
"model.llm.layers.7.attn.wv_dv.weight",
"model.llm.layers.7.attn_norm.weight",
"model.llm.layers.7.attn_sconv.weight",
"model.llm.layers.7.mlp.gate.weight",
"model.llm.layers.7.mlp_norm.weight",
"model.llm.layers.7.mlp_sconv.weight",
"model.llm.layers.8.attn.attn_sink_proj.weight",
"model.llm.layers.8.attn.gate_attn.weight",
"model.llm.layers.8.attn.k_sconv.weight",
"model.llm.layers.8.attn.q_l2_clip.scale",
"model.llm.layers.8.attn.rel_l2_clip.scale",
"model.llm.layers.8.attn.rel_logits_proj.proj",
"model.llm.layers.8.attn.v_sconv.weight",
"model.llm.layers.8.attn.wk_dv.weight",
"model.llm.layers.8.attn.wo_ud.weight",
"model.llm.layers.8.attn.wq_du.weight",
"model.llm.layers.8.attn.wr_du.weight",
"model.llm.layers.8.attn.wv_dv.weight",
"model.llm.layers.8.attn_norm.weight",
"model.llm.layers.8.attn_sconv.weight",
"model.llm.layers.8.mlp.gate.weight",
"model.llm.layers.8.mlp_norm.weight",
"model.llm.layers.8.mlp_sconv.weight",
"model.llm.layers.9.attn.attn_sink_proj.weight",
"model.llm.layers.9.attn.gate_attn.weight",
"model.llm.layers.9.attn.k_sconv.weight",
"model.llm.layers.9.attn.q_l2_clip.scale",
"model.llm.layers.9.attn.rel_l2_clip.scale",
"model.llm.layers.9.attn.rel_logits_proj.proj",
"model.llm.layers.9.attn.v_sconv.weight",
"model.llm.layers.9.attn.wk_dv.weight",
"model.llm.layers.9.attn.wo_ud.weight",
"model.llm.layers.9.attn.wq_du.weight",
"model.llm.layers.9.attn.wr_du.weight",
"model.llm.layers.9.attn.wv_dv.weight",
"model.llm.layers.9.attn_norm.weight",
"model.llm.layers.9.attn_sconv.weight",
"model.llm.layers.9.mlp.gate.weight",
"model.llm.layers.9.mlp_norm.weight",
"model.llm.layers.9.mlp_sconv.weight",
"model.llm.norm.weight",
"model.visual.backbone.embeddings.class_embedding",
"model.visual.backbone.embeddings.cls_position_embedding",
"model.visual.backbone.embeddings.patch_embedding.weight",
"model.visual.backbone.embeddings.position_embedding",
"model.visual.backbone.layers.0.ln_1.bias",
"model.visual.backbone.layers.0.ln_1.weight",
"model.visual.backbone.layers.0.ln_2.bias",
"model.visual.backbone.layers.0.ln_2.weight",
"model.visual.backbone.layers.0.ls_1.gamma",
"model.visual.backbone.layers.0.ls_2.gamma",
"model.visual.backbone.layers.0.mlp.c_fc.bias",
"model.visual.backbone.layers.0.mlp.c_fc.weight",
"model.visual.backbone.layers.0.mlp.c_proj.bias",
"model.visual.backbone.layers.0.mlp.c_proj.weight",
"model.visual.backbone.layers.0.self_attn.k_proj.bias",
"model.visual.backbone.layers.0.self_attn.k_proj.weight",
"model.visual.backbone.layers.0.self_attn.out_proj.bias",
"model.visual.backbone.layers.0.self_attn.out_proj.weight",
"model.visual.backbone.layers.0.self_attn.q_proj.bias",
"model.visual.backbone.layers.0.self_attn.q_proj.weight",
"model.visual.backbone.layers.0.self_attn.v_proj.bias",
"model.visual.backbone.layers.0.self_attn.v_proj.weight",
"model.visual.backbone.layers.1.ln_1.bias",
"model.visual.backbone.layers.1.ln_1.weight",
"model.visual.backbone.layers.1.ln_2.bias",
"model.visual.backbone.layers.1.ln_2.weight",
"model.visual.backbone.layers.1.ls_1.gamma",
"model.visual.backbone.layers.1.ls_2.gamma",
"model.visual.backbone.layers.1.mlp.c_fc.bias",
"model.visual.backbone.layers.1.mlp.c_fc.weight",
"model.visual.backbone.layers.1.mlp.c_proj.bias",
"model.visual.backbone.layers.1.mlp.c_proj.weight",
"model.visual.backbone.layers.1.self_attn.k_proj.bias",
"model.visual.backbone.layers.1.self_attn.k_proj.weight",
"model.visual.backbone.layers.1.self_attn.out_proj.bias",
"model.visual.backbone.layers.1.self_attn.out_proj.weight",
"model.visual.backbone.layers.1.self_attn.q_proj.bias",
"model.visual.backbone.layers.1.self_attn.q_proj.weight",
"model.visual.backbone.layers.1.self_attn.v_proj.bias",
"model.visual.backbone.layers.1.self_attn.v_proj.weight",
"model.visual.backbone.layers.10.ln_1.bias",
"model.visual.backbone.layers.10.ln_1.weight",
"model.visual.backbone.layers.10.ln_2.bias",
"model.visual.backbone.layers.10.ln_2.weight",
"model.visual.backbone.layers.10.ls_1.gamma",
"model.visual.backbone.layers.10.ls_2.gamma",
"model.visual.backbone.layers.10.mlp.c_fc.bias",
"model.visual.backbone.layers.10.mlp.c_fc.weight",
"model.visual.backbone.layers.10.mlp.c_proj.bias",
"model.visual.backbone.layers.10.mlp.c_proj.weight",
"model.visual.backbone.layers.10.self_attn.k_proj.bias",
"model.visual.backbone.layers.10.self_attn.k_proj.weight",
"model.visual.backbone.layers.10.self_attn.out_proj.bias",
"model.visual.backbone.layers.10.self_attn.out_proj.weight",
"model.visual.backbone.layers.10.self_attn.q_proj.bias",
"model.visual.backbone.layers.10.self_attn.q_proj.weight",
"model.visual.backbone.layers.10.self_attn.v_proj.bias",
"model.visual.backbone.layers.10.self_attn.v_proj.weight",
"model.visual.backbone.layers.11.ln_1.bias",
"model.visual.backbone.layers.11.ln_1.weight",
"model.visual.backbone.layers.11.ln_2.bias",
"model.visual.backbone.layers.11.ln_2.weight",
"model.visual.backbone.layers.11.ls_1.gamma",
"model.visual.backbone.layers.11.ls_2.gamma",
"model.visual.backbone.layers.11.mlp.c_fc.bias",
"model.visual.backbone.layers.11.mlp.c_fc.weight",
"model.visual.backbone.layers.11.mlp.c_proj.bias",
"model.visual.backbone.layers.11.mlp.c_proj.weight",
"model.visual.backbone.layers.11.self_attn.k_proj.bias",
"model.visual.backbone.layers.11.self_attn.k_proj.weight",
"model.visual.backbone.layers.11.self_attn.out_proj.bias",
"model.visual.backbone.layers.11.self_attn.out_proj.weight",
"model.visual.backbone.layers.11.self_attn.q_proj.bias",
"model.visual.backbone.layers.11.self_attn.q_proj.weight",
"model.visual.backbone.layers.11.self_attn.v_proj.bias",
"model.visual.backbone.layers.11.self_attn.v_proj.weight",
"model.visual.backbone.layers.12.ln_1.bias",
"model.visual.backbone.layers.12.ln_1.weight",
"model.visual.backbone.layers.12.ln_2.bias",
"model.visual.backbone.layers.12.ln_2.weight",
"model.visual.backbone.layers.12.ls_1.gamma",
"model.visual.backbone.layers.12.ls_2.gamma",
"model.visual.backbone.layers.12.mlp.c_fc.bias",
"model.visual.backbone.layers.12.mlp.c_fc.weight",
"model.visual.backbone.layers.12.mlp.c_proj.bias",
"model.visual.backbone.layers.12.mlp.c_proj.weight",
"model.visual.backbone.layers.12.self_attn.k_proj.bias",
"model.visual.backbone.layers.12.self_attn.k_proj.weight",
"model.visual.backbone.layers.12.self_attn.out_proj.bias",
"model.visual.backbone.layers.12.self_attn.out_proj.weight",
"model.visual.backbone.layers.12.self_attn.q_proj.bias",
"model.visual.backbone.layers.12.self_attn.q_proj.weight",
"model.visual.backbone.layers.12.self_attn.v_proj.bias",
"model.visual.backbone.layers.12.self_attn.v_proj.weight",
"model.visual.backbone.layers.13.ln_1.bias",
"model.visual.backbone.layers.13.ln_1.weight",
"model.visual.backbone.layers.13.ln_2.bias",
"model.visual.backbone.layers.13.ln_2.weight",
"model.visual.backbone.layers.13.ls_1.gamma",
"model.visual.backbone.layers.13.ls_2.gamma",
"model.visual.backbone.layers.13.mlp.c_fc.bias",
"model.visual.backbone.layers.13.mlp.c_fc.weight",
"model.visual.backbone.layers.13.mlp.c_proj.bias",
"model.visual.backbone.layers.13.mlp.c_proj.weight",
"model.visual.backbone.layers.13.self_attn.k_proj.bias",
"model.visual.backbone.layers.13.self_attn.k_proj.weight",
"model.visual.backbone.layers.13.self_attn.out_proj.bias",
"model.visual.backbone.layers.13.self_attn.out_proj.weight",
"model.visual.backbone.layers.13.self_attn.q_proj.bias",
"model.visual.backbone.layers.13.self_attn.q_proj.weight",
"model.visual.backbone.layers.13.self_attn.v_proj.bias",
"model.visual.backbone.layers.13.self_attn.v_proj.weight",
"model.visual.backbone.layers.14.ln_1.bias",
"model.visual.backbone.layers.14.ln_1.weight",
"model.visual.backbone.layers.14.ln_2.bias",
"model.visual.backbone.layers.14.ln_2.weight",
"model.visual.backbone.layers.14.ls_1.gamma",
"model.visual.backbone.layers.14.ls_2.gamma",
"model.visual.backbone.layers.14.mlp.c_fc.bias",
"model.visual.backbone.layers.14.mlp.c_fc.weight",
"model.visual.backbone.layers.14.mlp.c_proj.bias",
"model.visual.backbone.layers.14.mlp.c_proj.weight",
"model.visual.backbone.layers.14.self_attn.k_proj.bias",
"model.visual.backbone.layers.14.self_attn.k_proj.weight",
"model.visual.backbone.layers.14.self_attn.out_proj.bias",
"model.visual.backbone.layers.14.self_attn.out_proj.weight",
"model.visual.backbone.layers.14.self_attn.q_proj.bias",
"model.visual.backbone.layers.14.self_attn.q_proj.weight",
"model.visual.backbone.layers.14.self_attn.v_proj.bias",
"model.visual.backbone.layers.14.self_attn.v_proj.weight",
"model.visual.backbone.layers.15.ln_1.bias",
"model.visual.backbone.layers.15.ln_1.weight",
"model.visual.backbone.layers.15.ln_2.bias",
"model.visual.backbone.layers.15.ln_2.weight",
"model.visual.backbone.layers.15.ls_1.gamma",
"model.visual.backbone.layers.15.ls_2.gamma",
"model.visual.backbone.layers.15.mlp.c_fc.bias",
"model.visual.backbone.layers.15.mlp.c_fc.weight",
"model.visual.backbone.layers.15.mlp.c_proj.bias",
"model.visual.backbone.layers.15.mlp.c_proj.weight",
"model.visual.backbone.layers.15.self_attn.k_proj.bias",
"model.visual.backbone.layers.15.self_attn.k_proj.weight",
"model.visual.backbone.layers.15.self_attn.out_proj.bias",
"model.visual.backbone.layers.15.self_attn.out_proj.weight",
"model.visual.backbone.layers.15.self_attn.q_proj.bias",
"model.visual.backbone.layers.15.self_attn.q_proj.weight",
"model.visual.backbone.layers.15.self_attn.v_proj.bias",
"model.visual.backbone.layers.15.self_attn.v_proj.weight",
"model.visual.backbone.layers.16.ln_1.bias",
"model.visual.backbone.layers.16.ln_1.weight",
"model.visual.backbone.layers.16.ln_2.bias",
"model.visual.backbone.layers.16.ln_2.weight",
"model.visual.backbone.layers.16.ls_1.gamma",
"model.visual.backbone.layers.16.ls_2.gamma",
"model.visual.backbone.layers.16.mlp.c_fc.bias",
"model.visual.backbone.layers.16.mlp.c_fc.weight",
"model.visual.backbone.layers.16.mlp.c_proj.bias",
"model.visual.backbone.layers.16.mlp.c_proj.weight",
"model.visual.backbone.layers.16.self_attn.k_proj.bias",
"model.visual.backbone.layers.16.self_attn.k_proj.weight",
"model.visual.backbone.layers.16.self_attn.out_proj.bias",
"model.visual.backbone.layers.16.self_attn.out_proj.weight",
"model.visual.backbone.layers.16.self_attn.q_proj.bias",
"model.visual.backbone.layers.16.self_attn.q_proj.weight",
"model.visual.backbone.layers.16.self_attn.v_proj.bias",
"model.visual.backbone.layers.16.self_attn.v_proj.weight",
"model.visual.backbone.layers.17.ln_1.bias",
"model.visual.backbone.layers.17.ln_1.weight",
"model.visual.backbone.layers.17.ln_2.bias",
"model.visual.backbone.layers.17.ln_2.weight",
"model.visual.backbone.layers.17.ls_1.gamma",
"model.visual.backbone.layers.17.ls_2.gamma",
"model.visual.backbone.layers.17.mlp.c_fc.bias",
"model.visual.backbone.layers.17.mlp.c_fc.weight",
"model.visual.backbone.layers.17.mlp.c_proj.bias",
"model.visual.backbone.layers.17.mlp.c_proj.weight",
"model.visual.backbone.layers.17.self_attn.k_proj.bias",
"model.visual.backbone.layers.17.self_attn.k_proj.weight",
"model.visual.backbone.layers.17.self_attn.out_proj.bias",
"model.visual.backbone.layers.17.self_attn.out_proj.weight",
"model.visual.backbone.layers.17.self_attn.q_proj.bias",
"model.visual.backbone.layers.17.self_attn.q_proj.weight",
"model.visual.backbone.layers.17.self_attn.v_proj.bias",
"model.visual.backbone.layers.17.self_attn.v_proj.weight",
"model.visual.backbone.layers.18.ln_1.bias",
"model.visual.backbone.layers.18.ln_1.weight",
"model.visual.backbone.layers.18.ln_2.bias",
"model.visual.backbone.layers.18.ln_2.weight",
"model.visual.backbone.layers.18.ls_1.gamma",
"model.visual.backbone.layers.18.ls_2.gamma",
"model.visual.backbone.layers.18.mlp.c_fc.bias",
"model.visual.backbone.layers.18.mlp.c_fc.weight",
"model.visual.backbone.layers.18.mlp.c_proj.bias",
"model.visual.backbone.layers.18.mlp.c_proj.weight",
"model.visual.backbone.layers.18.self_attn.k_proj.bias",
"model.visual.backbone.layers.18.self_attn.k_proj.weight",
"model.visual.backbone.layers.18.self_attn.out_proj.bias",
"model.visual.backbone.layers.18.self_attn.out_proj.weight",
"model.visual.backbone.layers.18.self_attn.q_proj.bias",
"model.visual.backbone.layers.18.self_attn.q_proj.weight",
"model.visual.backbone.layers.18.self_attn.v_proj.bias",
"model.visual.backbone.layers.18.self_attn.v_proj.weight",
"model.visual.backbone.layers.19.ln_1.bias",
"model.visual.backbone.layers.19.ln_1.weight",
"model.visual.backbone.layers.19.ln_2.bias",
"model.visual.backbone.layers.19.ln_2.weight",
"model.visual.backbone.layers.19.ls_1.gamma",
"model.visual.backbone.layers.19.ls_2.gamma",
"model.visual.backbone.layers.19.mlp.c_fc.bias",
"model.visual.backbone.layers.19.mlp.c_fc.weight",
"model.visual.backbone.layers.19.mlp.c_proj.bias",
"model.visual.backbone.layers.19.mlp.c_proj.weight",
"model.visual.backbone.layers.19.self_attn.k_proj.bias",
"model.visual.backbone.layers.19.self_attn.k_proj.weight",
"model.visual.backbone.layers.19.self_attn.out_proj.bias",
"model.visual.backbone.layers.19.self_attn.out_proj.weight",
"model.visual.backbone.layers.19.self_attn.q_proj.bias",
"model.visual.backbone.layers.19.self_attn.q_proj.weight",
"model.visual.backbone.layers.19.self_attn.v_proj.bias",
"model.visual.backbone.layers.19.self_attn.v_proj.weight",
"model.visual.backbone.layers.2.ln_1.bias",
"model.visual.backbone.layers.2.ln_1.weight",
"model.visual.backbone.layers.2.ln_2.bias",
"model.visual.backbone.layers.2.ln_2.weight",
"model.visual.backbone.layers.2.ls_1.gamma",
"model.visual.backbone.layers.2.ls_2.gamma",
"model.visual.backbone.layers.2.mlp.c_fc.bias",
"model.visual.backbone.layers.2.mlp.c_fc.weight",
"model.visual.backbone.layers.2.mlp.c_proj.bias",
"model.visual.backbone.layers.2.mlp.c_proj.weight",
"model.visual.backbone.layers.2.self_attn.k_proj.bias",
"model.visual.backbone.layers.2.self_attn.k_proj.weight",
"model.visual.backbone.layers.2.self_attn.out_proj.bias",
"model.visual.backbone.layers.2.self_attn.out_proj.weight",
"model.visual.backbone.layers.2.self_attn.q_proj.bias",
"model.visual.backbone.layers.2.self_attn.q_proj.weight",
"model.visual.backbone.layers.2.self_attn.v_proj.bias",
"model.visual.backbone.layers.2.self_attn.v_proj.weight",
"model.visual.backbone.layers.20.ln_1.bias",
"model.visual.backbone.layers.20.ln_1.weight",
"model.visual.backbone.layers.20.ln_2.bias",
"model.visual.backbone.layers.20.ln_2.weight",
"model.visual.backbone.layers.20.ls_1.gamma",
"model.visual.backbone.layers.20.ls_2.gamma",
"model.visual.backbone.layers.20.mlp.c_fc.bias",
"model.visual.backbone.layers.20.mlp.c_fc.weight",
"model.visual.backbone.layers.20.mlp.c_proj.bias",
"model.visual.backbone.layers.20.mlp.c_proj.weight",
"model.visual.backbone.layers.20.self_attn.k_proj.bias",
"model.visual.backbone.layers.20.self_attn.k_proj.weight",
"model.visual.backbone.layers.20.self_attn.out_proj.bias",
"model.visual.backbone.layers.20.self_attn.out_proj.weight",
"model.visual.backbone.layers.20.self_attn.q_proj.bias",
"model.visual.backbone.layers.20.self_attn.q_proj.weight",
"model.visual.backbone.layers.20.self_attn.v_proj.bias",
"model.visual.backbone.layers.20.self_attn.v_proj.weight",
"model.visual.backbone.layers.21.ln_1.bias",
"model.visual.backbone.layers.21.ln_1.weight",
"model.visual.backbone.layers.21.ln_2.bias",
"model.visual.backbone.layers.21.ln_2.weight",
"model.visual.backbone.layers.21.ls_1.gamma",
"model.visual.backbone.layers.21.ls_2.gamma",
"model.visual.backbone.layers.21.mlp.c_fc.bias",
"model.visual.backbone.layers.21.mlp.c_fc.weight",
"model.visual.backbone.layers.21.mlp.c_proj.bias",
"model.visual.backbone.layers.21.mlp.c_proj.weight",
"model.visual.backbone.layers.21.self_attn.k_proj.bias",
"model.visual.backbone.layers.21.self_attn.k_proj.weight",
"model.visual.backbone.layers.21.self_attn.out_proj.bias",
"model.visual.backbone.layers.21.self_attn.out_proj.weight",
"model.visual.backbone.layers.21.self_attn.q_proj.bias",
"model.visual.backbone.layers.21.self_attn.q_proj.weight",
"model.visual.backbone.layers.21.self_attn.v_proj.bias",
"model.visual.backbone.layers.21.self_attn.v_proj.weight",
"model.visual.backbone.layers.22.ln_1.bias",
"model.visual.backbone.layers.22.ln_1.weight",
"model.visual.backbone.layers.22.ln_2.bias",
"model.visual.backbone.layers.22.ln_2.weight",
"model.visual.backbone.layers.22.ls_1.gamma",
"model.visual.backbone.layers.22.ls_2.gamma",
"model.visual.backbone.layers.22.mlp.c_fc.bias",
"model.visual.backbone.layers.22.mlp.c_fc.weight",
"model.visual.backbone.layers.22.mlp.c_proj.bias",
"model.visual.backbone.layers.22.mlp.c_proj.weight",
"model.visual.backbone.layers.22.self_attn.k_proj.bias",
"model.visual.backbone.layers.22.self_attn.k_proj.weight",
"model.visual.backbone.layers.22.self_attn.out_proj.bias",
"model.visual.backbone.layers.22.self_attn.out_proj.weight",
"model.visual.backbone.layers.22.self_attn.q_proj.bias",
"model.visual.backbone.layers.22.self_attn.q_proj.weight",
"model.visual.backbone.layers.22.self_attn.v_proj.bias",
"model.visual.backbone.layers.22.self_attn.v_proj.weight",
"model.visual.backbone.layers.3.ln_1.bias",
"model.visual.backbone.layers.3.ln_1.weight",
"model.visual.backbone.layers.3.ln_2.bias",
"model.visual.backbone.layers.3.ln_2.weight",
"model.visual.backbone.layers.3.ls_1.gamma",
"model.visual.backbone.layers.3.ls_2.gamma",
"model.visual.backbone.layers.3.mlp.c_fc.bias",
"model.visual.backbone.layers.3.mlp.c_fc.weight",
"model.visual.backbone.layers.3.mlp.c_proj.bias",
"model.visual.backbone.layers.3.mlp.c_proj.weight",
"model.visual.backbone.layers.3.self_attn.k_proj.bias",
"model.visual.backbone.layers.3.self_attn.k_proj.weight",
"model.visual.backbone.layers.3.self_attn.out_proj.bias",
"model.visual.backbone.layers.3.self_attn.out_proj.weight",
"model.visual.backbone.layers.3.self_attn.q_proj.bias",
"model.visual.backbone.layers.3.self_attn.q_proj.weight",
"model.visual.backbone.layers.3.self_attn.v_proj.bias",
"model.visual.backbone.layers.3.self_attn.v_proj.weight",
"model.visual.backbone.layers.4.ln_1.bias",
"model.visual.backbone.layers.4.ln_1.weight",
"model.visual.backbone.layers.4.ln_2.bias",
"model.visual.backbone.layers.4.ln_2.weight",
"model.visual.backbone.layers.4.ls_1.gamma",
"model.visual.backbone.layers.4.ls_2.gamma",
"model.visual.backbone.layers.4.mlp.c_fc.bias",
"model.visual.backbone.layers.4.mlp.c_fc.weight",
"model.visual.backbone.layers.4.mlp.c_proj.bias",
"model.visual.backbone.layers.4.mlp.c_proj.weight",
"model.visual.backbone.layers.4.self_attn.k_proj.bias",
"model.visual.backbone.layers.4.self_attn.k_proj.weight",
"model.visual.backbone.layers.4.self_attn.out_proj.bias",
"model.visual.backbone.layers.4.self_attn.out_proj.weight",
"model.visual.backbone.layers.4.self_attn.q_proj.bias",
"model.visual.backbone.layers.4.self_attn.q_proj.weight",
"model.visual.backbone.layers.4.self_attn.v_proj.bias",
"model.visual.backbone.layers.4.self_attn.v_proj.weight",
"model.visual.backbone.layers.5.ln_1.bias",
"model.visual.backbone.layers.5.ln_1.weight",
"model.visual.backbone.layers.5.ln_2.bias",
"model.visual.backbone.layers.5.ln_2.weight",
"model.visual.backbone.layers.5.ls_1.gamma",
"model.visual.backbone.layers.5.ls_2.gamma",
"model.visual.backbone.layers.5.mlp.c_fc.bias",
"model.visual.backbone.layers.5.mlp.c_fc.weight",
"model.visual.backbone.layers.5.mlp.c_proj.bias",
"model.visual.backbone.layers.5.mlp.c_proj.weight",
"model.visual.backbone.layers.5.self_attn.k_proj.bias",
"model.visual.backbone.layers.5.self_attn.k_proj.weight",
"model.visual.backbone.layers.5.self_attn.out_proj.bias",
"model.visual.backbone.layers.5.self_attn.out_proj.weight",
"model.visual.backbone.layers.5.self_attn.q_proj.bias",
"model.visual.backbone.layers.5.self_attn.q_proj.weight",
"model.visual.backbone.layers.5.self_attn.v_proj.bias",
"model.visual.backbone.layers.5.self_attn.v_proj.weight",
"model.visual.backbone.layers.6.ln_1.bias",
"model.visual.backbone.layers.6.ln_1.weight",
"model.visual.backbone.layers.6.ln_2.bias",
"model.visual.backbone.layers.6.ln_2.weight",
"model.visual.backbone.layers.6.ls_1.gamma",
"model.visual.backbone.layers.6.ls_2.gamma",
"model.visual.backbone.layers.6.mlp.c_fc.bias",
"model.visual.backbone.layers.6.mlp.c_fc.weight",
"model.visual.backbone.layers.6.mlp.c_proj.bias",
"model.visual.backbone.layers.6.mlp.c_proj.weight",
"model.visual.backbone.layers.6.self_attn.k_proj.bias",
"model.visual.backbone.layers.6.self_attn.k_proj.weight",
"model.visual.backbone.layers.6.self_attn.out_proj.bias",
"model.visual.backbone.layers.6.self_attn.out_proj.weight",
"model.visual.backbone.layers.6.self_attn.q_proj.bias",
"model.visual.backbone.layers.6.self_attn.q_proj.weight",
"model.visual.backbone.layers.6.self_attn.v_proj.bias",
"model.visual.backbone.layers.6.self_attn.v_proj.weight",
"model.visual.backbone.layers.7.ln_1.bias",
"model.visual.backbone.layers.7.ln_1.weight",
"model.visual.backbone.layers.7.ln_2.bias",
"model.visual.backbone.layers.7.ln_2.weight",
"model.visual.backbone.layers.7.ls_1.gamma",
"model.visual.backbone.layers.7.ls_2.gamma",
"model.visual.backbone.layers.7.mlp.c_fc.bias",
"model.visual.backbone.layers.7.mlp.c_fc.weight",
"model.visual.backbone.layers.7.mlp.c_proj.bias",
"model.visual.backbone.layers.7.mlp.c_proj.weight",
"model.visual.backbone.layers.7.self_attn.k_proj.bias",
"model.visual.backbone.layers.7.self_attn.k_proj.weight",
"model.visual.backbone.layers.7.self_attn.out_proj.bias",
"model.visual.backbone.layers.7.self_attn.out_proj.weight",
"model.visual.backbone.layers.7.self_attn.q_proj.bias",
"model.visual.backbone.layers.7.self_attn.q_proj.weight",
"model.visual.backbone.layers.7.self_attn.v_proj.bias",
"model.visual.backbone.layers.7.self_attn.v_proj.weight",
"model.visual.backbone.layers.8.ln_1.bias",
"model.visual.backbone.layers.8.ln_1.weight",
"model.visual.backbone.layers.8.ln_2.bias",
"model.visual.backbone.layers.8.ln_2.weight",
"model.visual.backbone.layers.8.ls_1.gamma",
"model.visual.backbone.layers.8.ls_2.gamma",
"model.visual.backbone.layers.8.mlp.c_fc.bias",
"model.visual.backbone.layers.8.mlp.c_fc.weight",
"model.visual.backbone.layers.8.mlp.c_proj.bias",
"model.visual.backbone.layers.8.mlp.c_proj.weight",
"model.visual.backbone.layers.8.self_attn.k_proj.bias",
"model.visual.backbone.layers.8.self_attn.k_proj.weight",
"model.visual.backbone.layers.8.self_attn.out_proj.bias",
"model.visual.backbone.layers.8.self_attn.out_proj.weight",
"model.visual.backbone.layers.8.self_attn.q_proj.bias",
"model.visual.backbone.layers.8.self_attn.q_proj.weight",
"model.visual.backbone.layers.8.self_attn.v_proj.bias",
"model.visual.backbone.layers.8.self_attn.v_proj.weight",
"model.visual.backbone.layers.9.ln_1.bias",
"model.visual.backbone.layers.9.ln_1.weight",
"model.visual.backbone.layers.9.ln_2.bias",
"model.visual.backbone.layers.9.ln_2.weight",
"model.visual.backbone.layers.9.ls_1.gamma",
"model.visual.backbone.layers.9.ls_2.gamma",
"model.visual.backbone.layers.9.mlp.c_fc.bias",
"model.visual.backbone.layers.9.mlp.c_fc.weight",
"model.visual.backbone.layers.9.mlp.c_proj.bias",
"model.visual.backbone.layers.9.mlp.c_proj.weight",
"model.visual.backbone.layers.9.self_attn.k_proj.bias",
"model.visual.backbone.layers.9.self_attn.k_proj.weight",
"model.visual.backbone.layers.9.self_attn.out_proj.bias",
"model.visual.backbone.layers.9.self_attn.out_proj.weight",
"model.visual.backbone.layers.9.self_attn.q_proj.bias",
"model.visual.backbone.layers.9.self_attn.q_proj.weight",
"model.visual.backbone.layers.9.self_attn.v_proj.bias",
"model.visual.backbone.layers.9.self_attn.v_proj.weight",
"model.visual.backbone.ln_pre.bias",
"model.visual.backbone.ln_pre.weight",
"model.visual.proj_in.weight",
"model.visual.proj_out.weight"
],
"weight_block_shape": [],
"weight_transform": "cast"
},
{
"compute_dtype": "nvfp4",
"output_dtype": "uint8",
"required_activation_quantization": {
"amax_type": "per_token",
"block_size": 16,
"format": "nvfp4_e2m1",
"rounding": "nearest",
"scale_dtype": "float8_e4m3fn",
"scale_selection_rule": "fouroversix_mae"
},
"tensors": [
"model.llm.layers.0.mlp.experts.w13_weight",
"model.llm.layers.0.mlp.experts.w2_weight",
"model.llm.layers.1.mlp.experts.w13_weight",
"model.llm.layers.1.mlp.experts.w2_weight",
"model.llm.layers.10.mlp.experts.w13_weight",
"model.llm.layers.10.mlp.experts.w2_weight",
"model.llm.layers.11.mlp.experts.w13_weight",
"model.llm.layers.11.mlp.experts.w2_weight",
"model.llm.layers.12.mlp.experts.w13_weight",
"model.llm.layers.12.mlp.experts.w2_weight",
"model.llm.layers.13.mlp.experts.w13_weight",
"model.llm.layers.13.mlp.experts.w2_weight",
"model.llm.layers.14.mlp.experts.w13_weight",
"model.llm.layers.14.mlp.experts.w2_weight",
"model.llm.layers.15.mlp.experts.w13_weight",
"model.llm.layers.15.mlp.experts.w2_weight",
"model.llm.layers.16.mlp.experts.w13_weight",
"model.llm.layers.16.mlp.experts.w2_weight",
"model.llm.layers.17.mlp.experts.w13_weight",
"model.llm.layers.17.mlp.experts.w2_weight",
"model.llm.layers.18.mlp.experts.w13_weight",
"model.llm.layers.18.mlp.experts.w2_weight",
"model.llm.layers.19.mlp.experts.w13_weight",
"model.llm.layers.19.mlp.experts.w2_weight",
"model.llm.layers.2.mlp.experts.w13_weight",
"model.llm.layers.2.mlp.experts.w2_weight",
"model.llm.layers.20.mlp.experts.w13_weight",
"model.llm.layers.20.mlp.experts.w2_weight",
"model.llm.layers.21.mlp.experts.w13_weight",
"model.llm.layers.21.mlp.experts.w2_weight",
"model.llm.layers.22.mlp.experts.w13_weight",
"model.llm.layers.22.mlp.experts.w2_weight",
"model.llm.layers.23.mlp.experts.w13_weight",
"model.llm.layers.23.mlp.experts.w2_weight",
"model.llm.layers.24.mlp.experts.w13_weight",
"model.llm.layers.24.mlp.experts.w2_weight",
"model.llm.layers.25.mlp.experts.w13_weight",
"model.llm.layers.25.mlp.experts.w2_weight",
"model.llm.layers.26.mlp.experts.w13_weight",
"model.llm.layers.26.mlp.experts.w2_weight",
"model.llm.layers.27.mlp.experts.w13_weight",
"model.llm.layers.27.mlp.experts.w2_weight",
"model.llm.layers.28.mlp.experts.w13_weight",
"model.llm.layers.28.mlp.experts.w2_weight",
"model.llm.layers.29.mlp.experts.w13_weight",
"model.llm.layers.29.mlp.experts.w2_weight",
"model.llm.layers.3.mlp.experts.w13_weight",
"model.llm.layers.3.mlp.experts.w2_weight",
"model.llm.layers.30.mlp.experts.w13_weight",
"model.llm.layers.30.mlp.experts.w2_weight",
"model.llm.layers.31.mlp.experts.w13_weight",
"model.llm.layers.31.mlp.experts.w2_weight",
"model.llm.layers.32.mlp.experts.w13_weight",
"model.llm.layers.32.mlp.experts.w2_weight",
"model.llm.layers.33.mlp.experts.w13_weight",
"model.llm.layers.33.mlp.experts.w2_weight",
"model.llm.layers.34.mlp.experts.w13_weight",
"model.llm.layers.34.mlp.experts.w2_weight",
"model.llm.layers.35.mlp.experts.w13_weight",
"model.llm.layers.35.mlp.experts.w2_weight",
"model.llm.layers.36.mlp.experts.w13_weight",
"model.llm.layers.36.mlp.experts.w2_weight",
"model.llm.layers.37.mlp.experts.w13_weight",
"model.llm.layers.37.mlp.experts.w2_weight",
"model.llm.layers.38.mlp.experts.w13_weight",
"model.llm.layers.38.mlp.experts.w2_weight",
"model.llm.layers.39.mlp.experts.w13_weight",
"model.llm.layers.39.mlp.experts.w2_weight",
"model.llm.layers.4.mlp.experts.w13_weight",
"model.llm.layers.4.mlp.experts.w2_weight",
"model.llm.layers.5.mlp.experts.w13_weight",
"model.llm.layers.5.mlp.experts.w2_weight",
"model.llm.layers.6.mlp.experts.w13_weight",
"model.llm.layers.6.mlp.experts.w2_weight",
"model.llm.layers.7.mlp.experts.w13_weight",
"model.llm.layers.7.mlp.experts.w2_weight",
"model.llm.layers.8.mlp.experts.w13_weight",
"model.llm.layers.8.mlp.experts.w2_weight",
"model.llm.layers.9.mlp.experts.w13_weight",
"model.llm.layers.9.mlp.experts.w2_weight"
],
"weight_block_shape": [
1,
16
],
"weight_scale_format": "amax_denominator",
"weight_transform": "nvfp4_nearest"
},
{
"compute_dtype": "mxfp8",
"output_dtype": "float8_e4m3fn",
"required_activation_quantization": {
"block_size": 32,
"format": "mxfp8_e4m3",
"rounding": "nearest",
"scale_dtype": "ue8m0"
},
"tensors": [
"model.llm.layers.40.mlp.experts.w13_weight",
"model.llm.layers.40.mlp.experts.w2_weight",
"model.llm.layers.41.mlp.experts.w13_weight",
"model.llm.layers.41.mlp.experts.w2_weight",
"model.llm.layers.42.mlp.experts.w13_weight",
"model.llm.layers.42.mlp.experts.w2_weight",
"model.llm.layers.43.mlp.experts.w13_weight",
"model.llm.layers.43.mlp.experts.w2_weight",
"model.llm.layers.44.mlp.experts.w13_weight",
"model.llm.layers.44.mlp.experts.w2_weight",
"model.llm.layers.45.mlp.experts.w13_weight",
"model.llm.layers.45.mlp.experts.w2_weight",
"model.llm.layers.46.mlp.experts.w13_weight",
"model.llm.layers.46.mlp.experts.w2_weight",
"model.llm.layers.47.mlp.experts.w13_weight",
"model.llm.layers.47.mlp.experts.w2_weight"
],
"weight_block_shape": [
32,
32
],
"weight_transform": "mxfp8_nearest"
},
{
"compute_dtype": "bfloat16",
"fwd_output_dtype": "float32",
"output_dtype": "bfloat16",
"required_activation_quantization": null,
"tensors": [
"model.llm.unembed.weight"
],
"weight_block_shape": [],
"weight_transform": "cast"
}
]
},
"mtp": {
"parameter_groups": [
{
"compute_dtype": "bfloat16",
"output_dtype": "bfloat16",
"required_activation_quantization": null,
"tensors": [
"fuse_weight",
"layers.0.dflare_norm.weight",
"layers.0.input_layernorm.weight",
"layers.0.mlp.down_proj.weight",
"layers.0.mlp.gate_proj.weight",
"layers.0.mlp.up_proj.weight",
"layers.0.post_attention_layernorm.weight",
"layers.0.self_attn.k_norm.weight",
"layers.0.self_attn.k_norm_ctx.weight",
"layers.0.self_attn.k_proj.weight",
"layers.0.self_attn.k_proj_ctx.weight",
"layers.0.self_attn.o_proj.weight",
"layers.0.self_attn.q_norm.weight",
"layers.0.self_attn.q_proj.weight",
"layers.0.self_attn.v_proj.weight",
"layers.0.self_attn.v_proj_ctx.weight",
"layers.1.dflare_norm.weight",
"layers.1.input_layernorm.weight",
"layers.1.mlp.down_proj.weight",
"layers.1.mlp.gate_proj.weight",
"layers.1.mlp.up_proj.weight",
"layers.1.post_attention_layernorm.weight",
"layers.1.self_attn.k_norm.weight",
"layers.1.self_attn.k_norm_ctx.weight",
"layers.1.self_attn.k_proj.weight",
"layers.1.self_attn.k_proj_ctx.weight",
"layers.1.self_attn.o_proj.weight",
"layers.1.self_attn.q_norm.weight",
"layers.1.self_attn.q_proj.weight",
"layers.1.self_attn.v_proj.weight",
"layers.1.self_attn.v_proj_ctx.weight",
"layers.2.dflare_norm.weight",
"layers.2.input_layernorm.weight",
"layers.2.mlp.down_proj.weight",
"layers.2.mlp.gate_proj.weight",
"layers.2.mlp.up_proj.weight",
"layers.2.post_attention_layernorm.weight",
"layers.2.self_attn.k_norm.weight",
"layers.2.self_attn.k_norm_ctx.weight",
"layers.2.self_attn.k_proj.weight",
"layers.2.self_attn.k_proj_ctx.weight",
"layers.2.self_attn.o_proj.weight",
"layers.2.self_attn.q_norm.weight",
"layers.2.self_attn.q_proj.weight",
"layers.2.self_attn.v_proj.weight",
"layers.2.self_attn.v_proj_ctx.weight",
"layers.3.dflare_norm.weight",
"layers.3.input_layernorm.weight",
"layers.3.mlp.down_proj.weight",
"layers.3.mlp.gate_proj.weight",
"layers.3.mlp.up_proj.weight",
"layers.3.post_attention_layernorm.weight",
"layers.3.self_attn.k_norm.weight",
"layers.3.self_attn.k_norm_ctx.weight",
"layers.3.self_attn.k_proj.weight",
"layers.3.self_attn.k_proj_ctx.weight",
"layers.3.self_attn.o_proj.weight",
"layers.3.self_attn.q_norm.weight",
"layers.3.self_attn.q_proj.weight",
"layers.3.self_attn.v_proj.weight",
"layers.3.self_attn.v_proj_ctx.weight",
"layers.4.dflare_norm.weight",
"layers.4.input_layernorm.weight",
"layers.4.mlp.down_proj.weight",
"layers.4.mlp.gate_proj.weight",
"layers.4.mlp.up_proj.weight",
"layers.4.post_attention_layernorm.weight",
"layers.4.self_attn.k_norm.weight",
"layers.4.self_attn.k_norm_ctx.weight",
"layers.4.self_attn.k_proj.weight",
"layers.4.self_attn.k_proj_ctx.weight",
"layers.4.self_attn.o_proj.weight",
"layers.4.self_attn.q_norm.weight",
"layers.4.self_attn.q_proj.weight",
"layers.4.self_attn.v_proj.weight",
"layers.4.self_attn.v_proj_ctx.weight",
"markov_head.dflare_norm.weight",
"markov_head.hidden_out.bias",
"markov_head.hidden_out.weight",
"markov_head.markov_w1.weight",
"markov_head.rnn_proj.bias",
"markov_head.rnn_proj.weight",
"mask_embed",
"norm.weight"
],
"weight_block_shape": [],
"weight_transform": "cast"
}
]
},
"quant_method": "inkling2"
}
{
"model": {
"parameter_groups": [
{
"compute_dtype": "float32",
"output_dtype": "float32",
"required_activation_quantization": null,
"tensors": [
"model.llm.attn_res.attn_queries.0",
"model.llm.attn_res.attn_queries.1",
"model.llm.attn_res.attn_queries.10",
"model.llm.attn_res.attn_queries.11",
"model.llm.attn_res.attn_queries.12",
"model.llm.attn_res.attn_queries.13",
"model.llm.attn_res.attn_queries.14",
"model.llm.attn_res.attn_queries.15",
"model.llm.attn_res.attn_queries.16",
"model.llm.attn_res.attn_queries.17",
"model.llm.attn_res.attn_queries.18",
"model.llm.attn_res.attn_queries.19",
"model.llm.attn_res.attn_queries.2",
"model.llm.attn_res.attn_queries.20",
"model.llm.attn_res.attn_queries.21",
"model.llm.attn_res.attn_queries.22",
"model.llm.attn_res.attn_queries.23",
"model.llm.attn_res.attn_queries.24",
"model.llm.attn_res.attn_queries.25",
"model.llm.attn_res.attn_queries.26",
"model.llm.attn_res.attn_queries.27",
"model.llm.attn_res.attn_queries.28",
"model.llm.attn_res.attn_queries.29",
"model.llm.attn_res.attn_queries.3",
"model.llm.attn_res.attn_queries.30",
"model.llm.attn_res.attn_queries.31",
"model.llm.attn_res.attn_queries.32",
"model.llm.attn_res.attn_queries.33",
"model.llm.attn_res.attn_queries.34",
"model.llm.attn_res.attn_queries.35",
"model.llm.attn_res.attn_queries.36",
"model.llm.attn_res.attn_queries.37",
"model.llm.attn_res.attn_queries.38",
"model.llm.attn_res.attn_queries.39",
"model.llm.attn_res.attn_queries.4",
"model.llm.attn_res.attn_queries.40",
"model.llm.attn_res.attn_queries.41",
"model.llm.attn_res.attn_queries.42",
"model.llm.attn_res.attn_queries.43",
"model.llm.attn_res.attn_queries.44",
"model.llm.attn_res.attn_queries.45",
"model.llm.attn_res.attn_queries.46",
"model.llm.attn_res.attn_queries.5",
"model.llm.attn_res.attn_queries.6",
"model.llm.attn_res.attn_queries.7",
"model.llm.attn_res.attn_queries.8",
"model.llm.attn_res.attn_queries.9",
"model.llm.attn_res.mlp_queries.0",
"model.llm.attn_res.mlp_queries.1",
"model.llm.attn_res.mlp_queries.10",
"model.llm.attn_res.mlp_queries.11",
"model.llm.attn_res.mlp_queries.12",
"model.llm.attn_res.mlp_queries.13",
"model.llm.attn_res.mlp_queries.14",
"model.llm.attn_res.mlp_queries.15",
"model.llm.attn_res.mlp_queries.16",
"model.llm.attn_res.mlp_queries.17",
"model.llm.attn_res.mlp_queries.18",
"model.llm.attn_res.mlp_queries.19",
"model.llm.attn_res.mlp_queries.2",
"model.llm.attn_res.mlp_queries.20",
"model.llm.attn_res.mlp_queries.21",
"model.llm.attn_res.mlp_queries.22",
"model.llm.attn_res.mlp_queries.23",
"model.llm.attn_res.mlp_queries.24",
"model.llm.attn_res.mlp_queries.25",
"model.llm.attn_res.mlp_queries.26",
"model.llm.attn_res.mlp_queries.27",
"model.llm.attn_res.mlp_queries.28",
"model.llm.attn_res.mlp_queries.29",
"model.llm.attn_res.mlp_queries.3",
"model.llm.attn_res.mlp_queries.30",
"model.llm.attn_res.mlp_queries.31",
"model.llm.attn_res.mlp_queries.32",
"model.llm.attn_res.mlp_queries.33",
"model.llm.attn_res.mlp_queries.34",
"model.llm.attn_res.mlp_queries.35",
"model.llm.attn_res.mlp_queries.36",
"model.llm.attn_res.mlp_queries.37",
"model.llm.attn_res.mlp_queries.38",
"model.llm.attn_res.mlp_queries.39",
"model.llm.attn_res.mlp_queries.4",
"model.llm.attn_res.mlp_queries.40",
"model.llm.attn_res.mlp_queries.41",
"model.llm.attn_res.mlp_queries.42",
"model.llm.attn_res.mlp_queries.43",
"model.llm.attn_res.mlp_queries.44",
"model.llm.attn_res.mlp_queries.45",
"model.llm.attn_res.mlp_queries.46",
"model.llm.attn_res.mlp_queries.47",
"model.llm.attn_res.mlp_queries.5",
"model.llm.attn_res.mlp_queries.6",