Qwen3-4B-Instruct-2507 Magnetron Snapshot
This repository contains a Magnetron snapshot converted from the original Hugging Face model Qwen/Qwen3-4B-Instruct-2507.
The snapshot is intended for inference with the Magnetron runtime. Weights are stored in mixed FP8 mode using float8_e4m3fn weight tensors and per-tensor float32 scale tensors, while activations/default runtime dtype are kept in bfloat16.
Model details
- Source model:
Qwen/Qwen3-4B-Instruct-2507 - Snapshot file:
qwen3-4b-instruct-2507-f8e4m3fn.mag - Magnetron dtype mode:
f8e4m3fn - Mixed FP8 weights:
True - Tensor count:
650
Qwen3 configuration
| Field | Value |
|---|---|
vocab_size |
151936 |
hidden_size |
2560 |
intermediate_size |
9728 |
num_hidden_layers |
36 |
num_attention_heads |
32 |
num_key_value_heads |
8 |
head_dim |
128 |
max_position_embeddings |
8192 |
rms_norm_eps |
1e-06 |
tie_word_embeddings |
True |
rope_theta |
5000000.0 |
sliding_window |
None |
bos_token_id |
151643 |
eos_token_id |
151645 |
sampling_strategy |
SamplingStrategy.GREEDY |
quant_dtype |
magnetron.dtype.float8_e4m3fn |
Tensor manifest
| Name | Shape | DType |
|---|---|---|
embed_tokens.weight |
151936x2560 |
bf16 |
layers.0.input_layernorm.weight |
2560 |
bf16 |
layers.0.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.0.mlp.down_proj.weight_scale |
1 |
f32 |
layers.0.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.0.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.0.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.0.mlp.up_proj.weight_scale |
1 |
f32 |
layers.0.post_attention_layernorm.weight |
2560 |
bf16 |
layers.0.self_attn.k_norm.weight |
128 |
bf16 |
layers.0.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.0.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.0.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.0.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.0.self_attn.q_norm.weight |
128 |
bf16 |
layers.0.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.0.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.0.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.0.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.1.input_layernorm.weight |
2560 |
bf16 |
layers.1.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.1.mlp.down_proj.weight_scale |
1 |
f32 |
layers.1.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.1.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.1.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.1.mlp.up_proj.weight_scale |
1 |
f32 |
layers.1.post_attention_layernorm.weight |
2560 |
bf16 |
layers.1.self_attn.k_norm.weight |
128 |
bf16 |
layers.1.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.1.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.1.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.1.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.1.self_attn.q_norm.weight |
128 |
bf16 |
layers.1.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.1.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.1.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.1.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.10.input_layernorm.weight |
2560 |
bf16 |
layers.10.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.10.mlp.down_proj.weight_scale |
1 |
f32 |
layers.10.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.10.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.10.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.10.mlp.up_proj.weight_scale |
1 |
f32 |
layers.10.post_attention_layernorm.weight |
2560 |
bf16 |
layers.10.self_attn.k_norm.weight |
128 |
bf16 |
layers.10.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.10.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.10.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.10.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.10.self_attn.q_norm.weight |
128 |
bf16 |
layers.10.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.10.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.10.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.10.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.11.input_layernorm.weight |
2560 |
bf16 |
layers.11.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.11.mlp.down_proj.weight_scale |
1 |
f32 |
layers.11.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.11.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.11.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.11.mlp.up_proj.weight_scale |
1 |
f32 |
layers.11.post_attention_layernorm.weight |
2560 |
bf16 |
layers.11.self_attn.k_norm.weight |
128 |
bf16 |
layers.11.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.11.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.11.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.11.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.11.self_attn.q_norm.weight |
128 |
bf16 |
layers.11.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.11.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.11.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.11.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.12.input_layernorm.weight |
2560 |
bf16 |
layers.12.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.12.mlp.down_proj.weight_scale |
1 |
f32 |
layers.12.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.12.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.12.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.12.mlp.up_proj.weight_scale |
1 |
f32 |
layers.12.post_attention_layernorm.weight |
2560 |
bf16 |
layers.12.self_attn.k_norm.weight |
128 |
bf16 |
layers.12.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.12.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.12.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.12.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.12.self_attn.q_norm.weight |
128 |
bf16 |
layers.12.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.12.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.12.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.12.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.13.input_layernorm.weight |
2560 |
bf16 |
layers.13.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.13.mlp.down_proj.weight_scale |
1 |
f32 |
layers.13.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.13.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.13.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.13.mlp.up_proj.weight_scale |
1 |
f32 |
layers.13.post_attention_layernorm.weight |
2560 |
bf16 |
layers.13.self_attn.k_norm.weight |
128 |
bf16 |
layers.13.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.13.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.13.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.13.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.13.self_attn.q_norm.weight |
128 |
bf16 |
layers.13.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.13.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.13.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.13.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.14.input_layernorm.weight |
2560 |
bf16 |
layers.14.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.14.mlp.down_proj.weight_scale |
1 |
f32 |
layers.14.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.14.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.14.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.14.mlp.up_proj.weight_scale |
1 |
f32 |
layers.14.post_attention_layernorm.weight |
2560 |
bf16 |
layers.14.self_attn.k_norm.weight |
128 |
bf16 |
layers.14.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.14.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.14.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.14.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.14.self_attn.q_norm.weight |
128 |
bf16 |
layers.14.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.14.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.14.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.14.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.15.input_layernorm.weight |
2560 |
bf16 |
layers.15.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.15.mlp.down_proj.weight_scale |
1 |
f32 |
layers.15.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.15.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.15.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.15.mlp.up_proj.weight_scale |
1 |
f32 |
layers.15.post_attention_layernorm.weight |
2560 |
bf16 |
layers.15.self_attn.k_norm.weight |
128 |
bf16 |
layers.15.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.15.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.15.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.15.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.15.self_attn.q_norm.weight |
128 |
bf16 |
layers.15.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.15.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.15.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.15.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.16.input_layernorm.weight |
2560 |
bf16 |
layers.16.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.16.mlp.down_proj.weight_scale |
1 |
f32 |
layers.16.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.16.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.16.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.16.mlp.up_proj.weight_scale |
1 |
f32 |
layers.16.post_attention_layernorm.weight |
2560 |
bf16 |
layers.16.self_attn.k_norm.weight |
128 |
bf16 |
layers.16.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.16.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.16.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.16.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.16.self_attn.q_norm.weight |
128 |
bf16 |
layers.16.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.16.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.16.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.16.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.17.input_layernorm.weight |
2560 |
bf16 |
layers.17.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.17.mlp.down_proj.weight_scale |
1 |
f32 |
layers.17.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.17.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.17.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.17.mlp.up_proj.weight_scale |
1 |
f32 |
layers.17.post_attention_layernorm.weight |
2560 |
bf16 |
layers.17.self_attn.k_norm.weight |
128 |
bf16 |
layers.17.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.17.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.17.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.17.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.17.self_attn.q_norm.weight |
128 |
bf16 |
layers.17.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.17.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.17.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.17.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.18.input_layernorm.weight |
2560 |
bf16 |
layers.18.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.18.mlp.down_proj.weight_scale |
1 |
f32 |
layers.18.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.18.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.18.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.18.mlp.up_proj.weight_scale |
1 |
f32 |
layers.18.post_attention_layernorm.weight |
2560 |
bf16 |
layers.18.self_attn.k_norm.weight |
128 |
bf16 |
layers.18.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.18.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.18.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.18.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.18.self_attn.q_norm.weight |
128 |
bf16 |
layers.18.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.18.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.18.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.18.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.19.input_layernorm.weight |
2560 |
bf16 |
layers.19.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.19.mlp.down_proj.weight_scale |
1 |
f32 |
layers.19.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.19.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.19.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.19.mlp.up_proj.weight_scale |
1 |
f32 |
layers.19.post_attention_layernorm.weight |
2560 |
bf16 |
layers.19.self_attn.k_norm.weight |
128 |
bf16 |
layers.19.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.19.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.19.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.19.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.19.self_attn.q_norm.weight |
128 |
bf16 |
layers.19.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.19.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.19.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.19.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.2.input_layernorm.weight |
2560 |
bf16 |
layers.2.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.2.mlp.down_proj.weight_scale |
1 |
f32 |
layers.2.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.2.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.2.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.2.mlp.up_proj.weight_scale |
1 |
f32 |
layers.2.post_attention_layernorm.weight |
2560 |
bf16 |
layers.2.self_attn.k_norm.weight |
128 |
bf16 |
layers.2.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.2.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.2.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.2.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.2.self_attn.q_norm.weight |
128 |
bf16 |
layers.2.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.2.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.2.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.2.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.20.input_layernorm.weight |
2560 |
bf16 |
layers.20.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.20.mlp.down_proj.weight_scale |
1 |
f32 |
layers.20.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.20.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.20.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.20.mlp.up_proj.weight_scale |
1 |
f32 |
layers.20.post_attention_layernorm.weight |
2560 |
bf16 |
layers.20.self_attn.k_norm.weight |
128 |
bf16 |
layers.20.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.20.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.20.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.20.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.20.self_attn.q_norm.weight |
128 |
bf16 |
layers.20.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.20.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.20.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.20.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.21.input_layernorm.weight |
2560 |
bf16 |
layers.21.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.21.mlp.down_proj.weight_scale |
1 |
f32 |
layers.21.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.21.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.21.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.21.mlp.up_proj.weight_scale |
1 |
f32 |
layers.21.post_attention_layernorm.weight |
2560 |
bf16 |
layers.21.self_attn.k_norm.weight |
128 |
bf16 |
layers.21.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.21.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.21.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.21.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.21.self_attn.q_norm.weight |
128 |
bf16 |
layers.21.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.21.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.21.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.21.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.22.input_layernorm.weight |
2560 |
bf16 |
layers.22.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.22.mlp.down_proj.weight_scale |
1 |
f32 |
layers.22.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.22.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.22.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.22.mlp.up_proj.weight_scale |
1 |
f32 |
layers.22.post_attention_layernorm.weight |
2560 |
bf16 |
layers.22.self_attn.k_norm.weight |
128 |
bf16 |
layers.22.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.22.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.22.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.22.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.22.self_attn.q_norm.weight |
128 |
bf16 |
layers.22.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.22.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.22.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.22.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.23.input_layernorm.weight |
2560 |
bf16 |
layers.23.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.23.mlp.down_proj.weight_scale |
1 |
f32 |
layers.23.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.23.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.23.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.23.mlp.up_proj.weight_scale |
1 |
f32 |
layers.23.post_attention_layernorm.weight |
2560 |
bf16 |
layers.23.self_attn.k_norm.weight |
128 |
bf16 |
layers.23.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.23.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.23.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.23.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.23.self_attn.q_norm.weight |
128 |
bf16 |
layers.23.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.23.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.23.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.23.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.24.input_layernorm.weight |
2560 |
bf16 |
layers.24.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.24.mlp.down_proj.weight_scale |
1 |
f32 |
layers.24.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.24.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.24.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.24.mlp.up_proj.weight_scale |
1 |
f32 |
layers.24.post_attention_layernorm.weight |
2560 |
bf16 |
layers.24.self_attn.k_norm.weight |
128 |
bf16 |
layers.24.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.24.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.24.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.24.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.24.self_attn.q_norm.weight |
128 |
bf16 |
layers.24.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.24.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.24.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.24.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.25.input_layernorm.weight |
2560 |
bf16 |
layers.25.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.25.mlp.down_proj.weight_scale |
1 |
f32 |
layers.25.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.25.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.25.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.25.mlp.up_proj.weight_scale |
1 |
f32 |
layers.25.post_attention_layernorm.weight |
2560 |
bf16 |
layers.25.self_attn.k_norm.weight |
128 |
bf16 |
layers.25.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.25.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.25.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.25.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.25.self_attn.q_norm.weight |
128 |
bf16 |
layers.25.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.25.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.25.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.25.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.26.input_layernorm.weight |
2560 |
bf16 |
layers.26.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.26.mlp.down_proj.weight_scale |
1 |
f32 |
layers.26.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.26.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.26.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.26.mlp.up_proj.weight_scale |
1 |
f32 |
layers.26.post_attention_layernorm.weight |
2560 |
bf16 |
layers.26.self_attn.k_norm.weight |
128 |
bf16 |
layers.26.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.26.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.26.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.26.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.26.self_attn.q_norm.weight |
128 |
bf16 |
layers.26.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.26.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.26.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.26.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.27.input_layernorm.weight |
2560 |
bf16 |
layers.27.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.27.mlp.down_proj.weight_scale |
1 |
f32 |
layers.27.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.27.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.27.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.27.mlp.up_proj.weight_scale |
1 |
f32 |
layers.27.post_attention_layernorm.weight |
2560 |
bf16 |
layers.27.self_attn.k_norm.weight |
128 |
bf16 |
layers.27.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.27.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.27.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.27.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.27.self_attn.q_norm.weight |
128 |
bf16 |
layers.27.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.27.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.27.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.27.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.28.input_layernorm.weight |
2560 |
bf16 |
layers.28.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.28.mlp.down_proj.weight_scale |
1 |
f32 |
layers.28.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.28.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.28.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.28.mlp.up_proj.weight_scale |
1 |
f32 |
layers.28.post_attention_layernorm.weight |
2560 |
bf16 |
layers.28.self_attn.k_norm.weight |
128 |
bf16 |
layers.28.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.28.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.28.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.28.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.28.self_attn.q_norm.weight |
128 |
bf16 |
layers.28.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.28.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.28.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.28.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.29.input_layernorm.weight |
2560 |
bf16 |
layers.29.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.29.mlp.down_proj.weight_scale |
1 |
f32 |
layers.29.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.29.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.29.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.29.mlp.up_proj.weight_scale |
1 |
f32 |
layers.29.post_attention_layernorm.weight |
2560 |
bf16 |
layers.29.self_attn.k_norm.weight |
128 |
bf16 |
layers.29.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.29.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.29.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.29.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.29.self_attn.q_norm.weight |
128 |
bf16 |
layers.29.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.29.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.29.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.29.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.3.input_layernorm.weight |
2560 |
bf16 |
layers.3.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.3.mlp.down_proj.weight_scale |
1 |
f32 |
layers.3.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.3.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.3.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.3.mlp.up_proj.weight_scale |
1 |
f32 |
layers.3.post_attention_layernorm.weight |
2560 |
bf16 |
layers.3.self_attn.k_norm.weight |
128 |
bf16 |
layers.3.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.3.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.3.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.3.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.3.self_attn.q_norm.weight |
128 |
bf16 |
layers.3.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.3.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.3.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.3.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.30.input_layernorm.weight |
2560 |
bf16 |
layers.30.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.30.mlp.down_proj.weight_scale |
1 |
f32 |
layers.30.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.30.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.30.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.30.mlp.up_proj.weight_scale |
1 |
f32 |
layers.30.post_attention_layernorm.weight |
2560 |
bf16 |
layers.30.self_attn.k_norm.weight |
128 |
bf16 |
layers.30.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.30.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.30.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.30.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.30.self_attn.q_norm.weight |
128 |
bf16 |
layers.30.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.30.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.30.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.30.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.31.input_layernorm.weight |
2560 |
bf16 |
layers.31.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.31.mlp.down_proj.weight_scale |
1 |
f32 |
layers.31.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.31.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.31.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.31.mlp.up_proj.weight_scale |
1 |
f32 |
layers.31.post_attention_layernorm.weight |
2560 |
bf16 |
layers.31.self_attn.k_norm.weight |
128 |
bf16 |
layers.31.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.31.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.31.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.31.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.31.self_attn.q_norm.weight |
128 |
bf16 |
layers.31.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.31.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.31.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.31.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.32.input_layernorm.weight |
2560 |
bf16 |
layers.32.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.32.mlp.down_proj.weight_scale |
1 |
f32 |
layers.32.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.32.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.32.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.32.mlp.up_proj.weight_scale |
1 |
f32 |
layers.32.post_attention_layernorm.weight |
2560 |
bf16 |
layers.32.self_attn.k_norm.weight |
128 |
bf16 |
layers.32.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.32.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.32.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.32.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.32.self_attn.q_norm.weight |
128 |
bf16 |
layers.32.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.32.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.32.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.32.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.33.input_layernorm.weight |
2560 |
bf16 |
layers.33.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.33.mlp.down_proj.weight_scale |
1 |
f32 |
layers.33.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.33.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.33.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.33.mlp.up_proj.weight_scale |
1 |
f32 |
layers.33.post_attention_layernorm.weight |
2560 |
bf16 |
layers.33.self_attn.k_norm.weight |
128 |
bf16 |
layers.33.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.33.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.33.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.33.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.33.self_attn.q_norm.weight |
128 |
bf16 |
layers.33.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.33.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.33.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.33.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.34.input_layernorm.weight |
2560 |
bf16 |
layers.34.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.34.mlp.down_proj.weight_scale |
1 |
f32 |
layers.34.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.34.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.34.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.34.mlp.up_proj.weight_scale |
1 |
f32 |
layers.34.post_attention_layernorm.weight |
2560 |
bf16 |
layers.34.self_attn.k_norm.weight |
128 |
bf16 |
layers.34.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.34.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.34.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.34.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.34.self_attn.q_norm.weight |
128 |
bf16 |
layers.34.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.34.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.34.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.34.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.35.input_layernorm.weight |
2560 |
bf16 |
layers.35.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.35.mlp.down_proj.weight_scale |
1 |
f32 |
layers.35.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.35.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.35.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.35.mlp.up_proj.weight_scale |
1 |
f32 |
layers.35.post_attention_layernorm.weight |
2560 |
bf16 |
layers.35.self_attn.k_norm.weight |
128 |
bf16 |
layers.35.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.35.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.35.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.35.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.35.self_attn.q_norm.weight |
128 |
bf16 |
layers.35.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.35.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.35.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.35.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.4.input_layernorm.weight |
2560 |
bf16 |
layers.4.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.4.mlp.down_proj.weight_scale |
1 |
f32 |
layers.4.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.4.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.4.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.4.mlp.up_proj.weight_scale |
1 |
f32 |
layers.4.post_attention_layernorm.weight |
2560 |
bf16 |
layers.4.self_attn.k_norm.weight |
128 |
bf16 |
layers.4.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.4.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.4.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.4.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.4.self_attn.q_norm.weight |
128 |
bf16 |
layers.4.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.4.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.4.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.4.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.5.input_layernorm.weight |
2560 |
bf16 |
layers.5.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.5.mlp.down_proj.weight_scale |
1 |
f32 |
layers.5.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.5.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.5.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.5.mlp.up_proj.weight_scale |
1 |
f32 |
layers.5.post_attention_layernorm.weight |
2560 |
bf16 |
layers.5.self_attn.k_norm.weight |
128 |
bf16 |
layers.5.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.5.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.5.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.5.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.5.self_attn.q_norm.weight |
128 |
bf16 |
layers.5.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.5.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.5.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.5.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.6.input_layernorm.weight |
2560 |
bf16 |
layers.6.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.6.mlp.down_proj.weight_scale |
1 |
f32 |
layers.6.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.6.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.6.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.6.mlp.up_proj.weight_scale |
1 |
f32 |
layers.6.post_attention_layernorm.weight |
2560 |
bf16 |
layers.6.self_attn.k_norm.weight |
128 |
bf16 |
layers.6.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.6.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.6.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.6.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.6.self_attn.q_norm.weight |
128 |
bf16 |
layers.6.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.6.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.6.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.6.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.7.input_layernorm.weight |
2560 |
bf16 |
layers.7.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.7.mlp.down_proj.weight_scale |
1 |
f32 |
layers.7.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.7.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.7.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.7.mlp.up_proj.weight_scale |
1 |
f32 |
layers.7.post_attention_layernorm.weight |
2560 |
bf16 |
layers.7.self_attn.k_norm.weight |
128 |
bf16 |
layers.7.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.7.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.7.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.7.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.7.self_attn.q_norm.weight |
128 |
bf16 |
layers.7.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.7.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.7.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.7.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.8.input_layernorm.weight |
2560 |
bf16 |
layers.8.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.8.mlp.down_proj.weight_scale |
1 |
f32 |
layers.8.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.8.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.8.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.8.mlp.up_proj.weight_scale |
1 |
f32 |
layers.8.post_attention_layernorm.weight |
2560 |
bf16 |
layers.8.self_attn.k_norm.weight |
128 |
bf16 |
layers.8.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.8.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.8.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.8.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.8.self_attn.q_norm.weight |
128 |
bf16 |
layers.8.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.8.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.8.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.8.self_attn.v_proj.weight_scale |
1 |
f32 |
layers.9.input_layernorm.weight |
2560 |
bf16 |
layers.9.mlp.down_proj.weight |
2560x9728 |
f8e4m3fn |
layers.9.mlp.down_proj.weight_scale |
1 |
f32 |
layers.9.mlp.gate_proj.weight |
9728x2560 |
f8e4m3fn |
layers.9.mlp.gate_proj.weight_scale |
1 |
f32 |
layers.9.mlp.up_proj.weight |
9728x2560 |
f8e4m3fn |
layers.9.mlp.up_proj.weight_scale |
1 |
f32 |
layers.9.post_attention_layernorm.weight |
2560 |
bf16 |
layers.9.self_attn.k_norm.weight |
128 |
bf16 |
layers.9.self_attn.k_proj.weight |
1024x2560 |
f8e4m3fn |
layers.9.self_attn.k_proj.weight_scale |
1 |
f32 |
layers.9.self_attn.o_proj.weight |
2560x4096 |
f8e4m3fn |
layers.9.self_attn.o_proj.weight_scale |
1 |
f32 |
layers.9.self_attn.q_norm.weight |
128 |
bf16 |
layers.9.self_attn.q_proj.weight |
4096x2560 |
f8e4m3fn |
layers.9.self_attn.q_proj.weight_scale |
1 |
f32 |
layers.9.self_attn.v_proj.weight |
1024x2560 |
f8e4m3fn |
layers.9.self_attn.v_proj.weight_scale |
1 |
f32 |
norm.weight |
2560 |
bf16 |
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for mario-sieg/qwen3.0-4b-2507-instruct-magnetron
Base model
Qwen/Qwen3-4B-Instruct-2507