The current, most capable model that runs on a single GPU.
vision
1b
4b
12b
2,375 Pulls Updated 6 days ago
96b122408694 · 7.8GB
-
general.architecturegemma3gemma3
-
general.file_typeF16F16
-
gemma3.attention.head_count88
-
gemma3.attention.head_count_kv44
-
gemma3.attention.key_length256256
-
gemma3.attention.layer_norm_rms_epsilon1e-061e-06
-
gemma3.attention.sliding_window10241024
-
gemma3.attention.value_length256256
-
gemma3.block_count3434
-
gemma3.context_length131072131072
-
gemma3.embedding_length25602560
-
gemma3.feed_forward_length1024010240
-
gemma3.rope.freq_base1e+061e+06
-
gemma3.rope.scaling.factor88
-
gemma3.rope.scaling.typelinearlinear
-
tokenizer.ggml.add_bos_tokentruetrue
-
tokenizer.ggml.add_eos_tokenfalsefalse
-
tokenizer.ggml.add_space_prefixfalsefalse
-
tokenizer.ggml.bos_token_id22
-
tokenizer.ggml.eos_token_id11
-
tokenizer.ggml.modelllamallama
-
tokenizer.ggml.padding_token_id11
-
tokenizer.ggml.predefaultdefault
-
tokenizer.ggml.scores[-1000, -1000, -1000, -1000, -1000, ...][-1000, -1000, -1000, -1000, -1000, ...]
-
tokenizer.ggml.token_type[3, 3, 3, 3, 3, ...][3, 3, 3, 3, 3, ...]
-
tokenizer.ggml.tokens[<pad>, <eos>, <bos>, <unk>, <mask>, ...][<pad>, <eos>, <bos>, <unk>, <mask>, ...]
-
tokenizer.ggml.unknown_token_id33
-
token_embd.weightF16[2560, 262144]
-
blk.0.attn_k.weightF16[2560, 1024]
-
blk.0.attn_k_norm.weightF32[256]
-
blk.0.attn_norm.weightF32[2560]
-
blk.0.attn_output.weightF16[2048, 2560]
-
blk.0.attn_q.weightF16[2560, 2048]
-
blk.0.attn_q_norm.weightF32[256]
-
blk.0.attn_v.weightF16[2560, 1024]
-
blk.0.ffn_down.weightF16[10240, 2560]
-
blk.0.ffn_gate.weightF16[2560, 10240]
-
blk.0.ffn_norm.weightF32[2560]
-
blk.0.ffn_up.weightF16[2560, 10240]
-
blk.0.post_attention_norm.weightF32[2560]
-
blk.0.post_ffw_norm.weightF32[2560]
-
blk.1.attn_k.weightF16[2560, 1024]
-
blk.1.attn_k_norm.weightF32[256]
-
blk.1.attn_norm.weightF32[2560]
-
blk.1.attn_output.weightF16[2048, 2560]
-
blk.1.attn_q.weightF16[2560, 2048]
-
blk.1.attn_q_norm.weightF32[256]
-
blk.1.attn_v.weightF16[2560, 1024]
-
blk.1.ffn_down.weightF16[10240, 2560]
-
blk.1.ffn_gate.weightF16[2560, 10240]
-
blk.1.ffn_norm.weightF32[2560]
-
blk.1.ffn_up.weightF16[2560, 10240]
-
blk.1.post_attention_norm.weightF32[2560]
-
blk.1.post_ffw_norm.weightF32[2560]
-
blk.2.attn_k.weightF16[2560, 1024]
-
blk.2.attn_k_norm.weightF32[256]
-
blk.2.attn_norm.weightF32[2560]
-
blk.2.attn_output.weightF16[2048, 2560]
-
blk.2.attn_q.weightF16[2560, 2048]
-
blk.2.attn_q_norm.weightF32[256]
-
blk.2.attn_v.weightF16[2560, 1024]
-
blk.2.ffn_down.weightF16[10240, 2560]
-
blk.2.ffn_gate.weightF16[2560, 10240]
-
blk.2.ffn_norm.weightF32[2560]
-
blk.2.ffn_up.weightF16[2560, 10240]
-
blk.2.post_attention_norm.weightF32[2560]
-
blk.2.post_ffw_norm.weightF32[2560]
-
blk.3.attn_k.weightF16[2560, 1024]
-
blk.3.attn_k_norm.weightF32[256]
-
blk.3.attn_norm.weightF32[2560]
-
blk.3.attn_output.weightF16[2048, 2560]
-
blk.3.attn_q.weightF16[2560, 2048]
-
blk.3.attn_q_norm.weightF32[256]
-
blk.3.attn_v.weightF16[2560, 1024]
-
blk.3.ffn_down.weightF16[10240, 2560]
-
blk.3.ffn_gate.weightF16[2560, 10240]
-
blk.3.ffn_norm.weightF32[2560]
-
blk.3.ffn_up.weightF16[2560, 10240]
-
blk.3.post_attention_norm.weightF32[2560]
-
blk.3.post_ffw_norm.weightF32[2560]
-
blk.4.attn_k.weightF16[2560, 1024]
-
blk.4.attn_k_norm.weightF32[256]
-
blk.4.attn_norm.weightF32[2560]
-
blk.4.attn_output.weightF16[2048, 2560]
-
blk.4.attn_q.weightF16[2560, 2048]
-
blk.4.attn_q_norm.weightF32[256]
-
blk.4.attn_v.weightF16[2560, 1024]
-
blk.4.ffn_down.weightF16[10240, 2560]
-
blk.4.ffn_gate.weightF16[2560, 10240]
-
blk.4.ffn_norm.weightF32[2560]
-
blk.4.ffn_up.weightF16[2560, 10240]
-
blk.4.post_attention_norm.weightF32[2560]
-
blk.4.post_ffw_norm.weightF32[2560]
-
blk.5.attn_k.weightF16[2560, 1024]
-
blk.5.attn_k_norm.weightF32[256]
-
blk.5.attn_norm.weightF32[2560]
-
blk.5.attn_output.weightF16[2048, 2560]
-
blk.5.attn_q.weightF16[2560, 2048]
-
blk.5.attn_q_norm.weightF32[256]
-
blk.5.attn_v.weightF16[2560, 1024]
-
blk.5.ffn_down.weightF16[10240, 2560]
-
blk.5.ffn_gate.weightF16[2560, 10240]
-
blk.5.ffn_norm.weightF32[2560]
-
blk.5.ffn_up.weightF16[2560, 10240]
-
blk.5.post_attention_norm.weightF32[2560]
-
blk.5.post_ffw_norm.weightF32[2560]
-
blk.6.attn_k.weightF16[2560, 1024]
-
blk.6.attn_k_norm.weightF32[256]
-
blk.6.attn_norm.weightF32[2560]
-
blk.6.attn_output.weightF16[2048, 2560]
-
blk.6.attn_q.weightF16[2560, 2048]
-
blk.6.attn_q_norm.weightF32[256]
-
blk.6.attn_v.weightF16[2560, 1024]
-
blk.6.ffn_down.weightF16[10240, 2560]
-
blk.6.ffn_gate.weightF16[2560, 10240]
-
blk.6.ffn_norm.weightF32[2560]
-
blk.6.ffn_up.weightF16[2560, 10240]
-
blk.6.post_attention_norm.weightF32[2560]
-
blk.6.post_ffw_norm.weightF32[2560]
-
blk.7.attn_k.weightF16[2560, 1024]
-
blk.7.attn_k_norm.weightF32[256]
-
blk.7.attn_norm.weightF32[2560]
-
blk.7.attn_output.weightF16[2048, 2560]
-
blk.7.attn_q.weightF16[2560, 2048]
-
blk.7.attn_q_norm.weightF32[256]
-
blk.7.attn_v.weightF16[2560, 1024]
-
blk.7.ffn_down.weightF16[10240, 2560]
-
blk.7.ffn_gate.weightF16[2560, 10240]
-
blk.7.ffn_norm.weightF32[2560]
-
blk.7.ffn_up.weightF16[2560, 10240]
-
blk.7.post_attention_norm.weightF32[2560]
-
blk.7.post_ffw_norm.weightF32[2560]
-
blk.8.attn_k.weightF16[2560, 1024]
-
blk.8.attn_k_norm.weightF32[256]
-
blk.8.attn_norm.weightF32[2560]
-
blk.8.attn_output.weightF16[2048, 2560]
-
blk.8.attn_q.weightF16[2560, 2048]
-
blk.8.attn_q_norm.weightF32[256]
-
blk.8.attn_v.weightF16[2560, 1024]
-
blk.8.ffn_down.weightF16[10240, 2560]
-
blk.8.ffn_gate.weightF16[2560, 10240]
-
blk.8.ffn_norm.weightF32[2560]
-
blk.8.ffn_up.weightF16[2560, 10240]
-
blk.8.post_attention_norm.weightF32[2560]
-
blk.8.post_ffw_norm.weightF32[2560]
-
blk.9.attn_k.weightF16[2560, 1024]
-
blk.9.attn_k_norm.weightF32[256]
-
blk.9.attn_norm.weightF32[2560]
-
blk.9.attn_output.weightF16[2048, 2560]
-
blk.9.attn_q.weightF16[2560, 2048]
-
blk.9.attn_q_norm.weightF32[256]
-
blk.9.attn_v.weightF16[2560, 1024]
-
blk.9.ffn_down.weightF16[10240, 2560]
-
blk.9.ffn_gate.weightF16[2560, 10240]
-
blk.9.ffn_norm.weightF32[2560]
-
blk.9.ffn_up.weightF16[2560, 10240]
-
blk.9.post_attention_norm.weightF32[2560]
-
blk.9.post_ffw_norm.weightF32[2560]
-
blk.10.attn_k.weightF16[2560, 1024]
-
blk.10.attn_k_norm.weightF32[256]
-
blk.10.attn_norm.weightF32[2560]
-
blk.10.attn_output.weightF16[2048, 2560]
-
blk.10.attn_q.weightF16[2560, 2048]
-
blk.10.attn_q_norm.weightF32[256]
-
blk.10.attn_v.weightF16[2560, 1024]
-
blk.10.ffn_down.weightF16[10240, 2560]
-
blk.10.ffn_gate.weightF16[2560, 10240]
-
blk.10.ffn_norm.weightF32[2560]
-
blk.10.ffn_up.weightF16[2560, 10240]
-
blk.10.post_attention_norm.weightF32[2560]
-
blk.10.post_ffw_norm.weightF32[2560]
-
blk.11.attn_k.weightF16[2560, 1024]
-
blk.11.attn_k_norm.weightF32[256]
-
blk.11.attn_norm.weightF32[2560]
-
blk.11.attn_output.weightF16[2048, 2560]
-
blk.11.attn_q.weightF16[2560, 2048]
-
blk.11.attn_q_norm.weightF32[256]
-
blk.11.attn_v.weightF16[2560, 1024]
-
blk.11.ffn_down.weightF16[10240, 2560]
-
blk.11.ffn_gate.weightF16[2560, 10240]
-
blk.11.ffn_norm.weightF32[2560]
-
blk.11.ffn_up.weightF16[2560, 10240]
-
blk.11.post_attention_norm.weightF32[2560]
-
blk.11.post_ffw_norm.weightF32[2560]
-
blk.12.attn_k.weightF16[2560, 1024]
-
blk.12.attn_k_norm.weightF32[256]
-
blk.12.attn_norm.weightF32[2560]
-
blk.12.attn_output.weightF16[2048, 2560]
-
blk.12.attn_q.weightF16[2560, 2048]
-
blk.12.attn_q_norm.weightF32[256]
-
blk.12.attn_v.weightF16[2560, 1024]
-
blk.12.ffn_down.weightF16[10240, 2560]
-
blk.12.ffn_gate.weightF16[2560, 10240]
-
blk.12.ffn_norm.weightF32[2560]
-
blk.12.ffn_up.weightF16[2560, 10240]
-
blk.12.post_attention_norm.weightF32[2560]
-
blk.12.post_ffw_norm.weightF32[2560]
-
blk.13.attn_k.weightF16[2560, 1024]
-
blk.13.attn_k_norm.weightF32[256]
-
blk.13.attn_norm.weightF32[2560]
-
blk.13.attn_output.weightF16[2048, 2560]
-
blk.13.attn_q.weightF16[2560, 2048]
-
blk.13.attn_q_norm.weightF32[256]
-
blk.13.attn_v.weightF16[2560, 1024]
-
blk.13.ffn_down.weightF16[10240, 2560]
-
blk.13.ffn_gate.weightF16[2560, 10240]
-
blk.13.ffn_norm.weightF32[2560]
-
blk.13.ffn_up.weightF16[2560, 10240]
-
blk.13.post_attention_norm.weightF32[2560]
-
blk.13.post_ffw_norm.weightF32[2560]
-
blk.14.attn_k.weightF16[2560, 1024]
-
blk.14.attn_k_norm.weightF32[256]
-
blk.14.attn_norm.weightF32[2560]
-
blk.14.attn_output.weightF16[2048, 2560]
-
blk.14.attn_q.weightF16[2560, 2048]
-
blk.14.attn_q_norm.weightF32[256]
-
blk.14.attn_v.weightF16[2560, 1024]
-
blk.14.ffn_down.weightF16[10240, 2560]
-
blk.14.ffn_gate.weightF16[2560, 10240]
-
blk.14.ffn_norm.weightF32[2560]
-
blk.14.ffn_up.weightF16[2560, 10240]
-
blk.14.post_attention_norm.weightF32[2560]
-
blk.14.post_ffw_norm.weightF32[2560]
-
blk.15.attn_k.weightF16[2560, 1024]
-
blk.15.attn_k_norm.weightF32[256]
-
blk.15.attn_norm.weightF32[2560]
-
blk.15.attn_output.weightF16[2048, 2560]
-
blk.15.attn_q.weightF16[2560, 2048]
-
blk.15.attn_q_norm.weightF32[256]
-
blk.15.attn_v.weightF16[2560, 1024]
-
blk.15.ffn_down.weightF16[10240, 2560]
-
blk.15.ffn_gate.weightF16[2560, 10240]
-
blk.15.ffn_norm.weightF32[2560]
-
blk.15.ffn_up.weightF16[2560, 10240]
-
blk.15.post_attention_norm.weightF32[2560]
-
blk.15.post_ffw_norm.weightF32[2560]
-
blk.16.attn_k.weightF16[2560, 1024]
-
blk.16.attn_k_norm.weightF32[256]
-
blk.16.attn_norm.weightF32[2560]
-
blk.16.attn_output.weightF16[2048, 2560]
-
blk.16.attn_q.weightF16[2560, 2048]
-
blk.16.attn_q_norm.weightF32[256]
-
blk.16.attn_v.weightF16[2560, 1024]
-
blk.16.ffn_down.weightF16[10240, 2560]
-
blk.16.ffn_gate.weightF16[2560, 10240]
-
blk.16.ffn_norm.weightF32[2560]
-
blk.16.ffn_up.weightF16[2560, 10240]
-
blk.16.post_attention_norm.weightF32[2560]
-
blk.16.post_ffw_norm.weightF32[2560]
-
blk.17.attn_k.weightF16[2560, 1024]
-
blk.17.attn_k_norm.weightF32[256]
-
blk.17.attn_norm.weightF32[2560]
-
blk.17.attn_output.weightF16[2048, 2560]
-
blk.17.attn_q.weightF16[2560, 2048]
-
blk.17.attn_q_norm.weightF32[256]
-
blk.17.attn_v.weightF16[2560, 1024]
-
blk.17.ffn_down.weightF16[10240, 2560]
-
blk.17.ffn_gate.weightF16[2560, 10240]
-
blk.17.ffn_norm.weightF32[2560]
-
blk.17.ffn_up.weightF16[2560, 10240]
-
blk.17.post_attention_norm.weightF32[2560]
-
blk.17.post_ffw_norm.weightF32[2560]
-
blk.18.attn_k.weightF16[2560, 1024]
-
blk.18.attn_k_norm.weightF32[256]
-
blk.18.attn_norm.weightF32[2560]
-
blk.18.attn_output.weightF16[2048, 2560]
-
blk.18.attn_q.weightF16[2560, 2048]
-
blk.18.attn_q_norm.weightF32[256]
-
blk.18.attn_v.weightF16[2560, 1024]
-
blk.18.ffn_down.weightF16[10240, 2560]
-
blk.18.ffn_gate.weightF16[2560, 10240]
-
blk.18.ffn_norm.weightF32[2560]
-
blk.18.ffn_up.weightF16[2560, 10240]
-
blk.18.post_attention_norm.weightF32[2560]
-
blk.18.post_ffw_norm.weightF32[2560]
-
blk.19.attn_k.weightF16[2560, 1024]
-
blk.19.attn_k_norm.weightF32[256]
-
blk.19.attn_norm.weightF32[2560]
-
blk.19.attn_output.weightF16[2048, 2560]
-
blk.19.attn_q.weightF16[2560, 2048]
-
blk.19.attn_q_norm.weightF32[256]
-
blk.19.attn_v.weightF16[2560, 1024]
-
blk.19.ffn_down.weightF16[10240, 2560]
-
blk.19.ffn_gate.weightF16[2560, 10240]
-
blk.19.ffn_norm.weightF32[2560]
-
blk.19.ffn_up.weightF16[2560, 10240]
-
blk.19.post_attention_norm.weightF32[2560]
-
blk.19.post_ffw_norm.weightF32[2560]
-
blk.20.attn_k.weightF16[2560, 1024]
-
blk.20.attn_k_norm.weightF32[256]
-
blk.20.attn_norm.weightF32[2560]
-
blk.20.attn_output.weightF16[2048, 2560]
-
blk.20.attn_q.weightF16[2560, 2048]
-
blk.20.attn_q_norm.weightF32[256]
-
blk.20.attn_v.weightF16[2560, 1024]
-
blk.20.ffn_down.weightF16[10240, 2560]
-
blk.20.ffn_gate.weightF16[2560, 10240]
-
blk.20.ffn_norm.weightF32[2560]
-
blk.20.ffn_up.weightF16[2560, 10240]
-
blk.20.post_attention_norm.weightF32[2560]
-
blk.20.post_ffw_norm.weightF32[2560]
-
blk.21.attn_k.weightF16[2560, 1024]
-
blk.21.attn_k_norm.weightF32[256]
-
blk.21.attn_norm.weightF32[2560]
-
blk.21.attn_output.weightF16[2048, 2560]
-
blk.21.attn_q.weightF16[2560, 2048]
-
blk.21.attn_q_norm.weightF32[256]
-
blk.21.attn_v.weightF16[2560, 1024]
-
blk.21.ffn_down.weightF16[10240, 2560]
-
blk.21.ffn_gate.weightF16[2560, 10240]
-
blk.21.ffn_norm.weightF32[2560]
-
blk.21.ffn_up.weightF16[2560, 10240]
-
blk.21.post_attention_norm.weightF32[2560]
-
blk.21.post_ffw_norm.weightF32[2560]
-
blk.22.attn_k.weightF16[2560, 1024]
-
blk.22.attn_k_norm.weightF32[256]
-
blk.22.attn_norm.weightF32[2560]
-
blk.22.attn_output.weightF16[2048, 2560]
-
blk.22.attn_q.weightF16[2560, 2048]
-
blk.22.attn_q_norm.weightF32[256]
-
blk.22.attn_v.weightF16[2560, 1024]
-
blk.22.ffn_down.weightF16[10240, 2560]
-
blk.22.ffn_gate.weightF16[2560, 10240]
-
blk.22.ffn_norm.weightF32[2560]
-
blk.22.ffn_up.weightF16[2560, 10240]
-
blk.22.post_attention_norm.weightF32[2560]
-
blk.22.post_ffw_norm.weightF32[2560]
-
blk.23.attn_k.weightF16[2560, 1024]
-
blk.23.attn_k_norm.weightF32[256]
-
blk.23.attn_norm.weightF32[2560]
-
blk.23.attn_output.weightF16[2048, 2560]
-
blk.23.attn_q.weightF16[2560, 2048]
-
blk.23.attn_q_norm.weightF32[256]
-
blk.23.attn_v.weightF16[2560, 1024]
-
blk.23.ffn_down.weightF16[10240, 2560]
-
blk.23.ffn_gate.weightF16[2560, 10240]
-
blk.23.ffn_norm.weightF32[2560]
-
blk.23.ffn_up.weightF16[2560, 10240]
-
blk.23.post_attention_norm.weightF32[2560]
-
blk.23.post_ffw_norm.weightF32[2560]
-
blk.24.attn_k.weightF16[2560, 1024]
-
blk.24.attn_k_norm.weightF32[256]
-
blk.24.attn_norm.weightF32[2560]
-
blk.24.attn_output.weightF16[2048, 2560]
-
blk.24.attn_q.weightF16[2560, 2048]
-
blk.24.attn_q_norm.weightF32[256]
-
blk.24.attn_v.weightF16[2560, 1024]
-
blk.24.ffn_down.weightF16[10240, 2560]
-
blk.24.ffn_gate.weightF16[2560, 10240]
-
blk.24.ffn_norm.weightF32[2560]
-
blk.24.ffn_up.weightF16[2560, 10240]
-
blk.24.post_attention_norm.weightF32[2560]
-
blk.24.post_ffw_norm.weightF32[2560]
-
blk.25.attn_k.weightF16[2560, 1024]
-
blk.25.attn_k_norm.weightF32[256]
-
blk.25.attn_norm.weightF32[2560]
-
blk.25.attn_output.weightF16[2048, 2560]
-
blk.25.attn_q.weightF16[2560, 2048]
-
blk.25.attn_q_norm.weightF32[256]
-
blk.25.attn_v.weightF16[2560, 1024]
-
blk.25.ffn_down.weightF16[10240, 2560]
-
blk.25.ffn_gate.weightF16[2560, 10240]
-
blk.25.ffn_norm.weightF32[2560]
-
blk.25.ffn_up.weightF16[2560, 10240]
-
blk.25.post_attention_norm.weightF32[2560]
-
blk.25.post_ffw_norm.weightF32[2560]
-
blk.26.attn_k.weightF16[2560, 1024]
-
blk.26.attn_k_norm.weightF32[256]
-
blk.26.attn_norm.weightF32[2560]
-
blk.26.attn_output.weightF16[2048, 2560]
-
blk.26.attn_q.weightF16[2560, 2048]
-
blk.26.attn_q_norm.weightF32[256]
-
blk.26.attn_v.weightF16[2560, 1024]
-
blk.26.ffn_down.weightF16[10240, 2560]
-
blk.26.ffn_gate.weightF16[2560, 10240]
-
blk.26.ffn_norm.weightF32[2560]
-
blk.26.ffn_up.weightF16[2560, 10240]
-
blk.26.post_attention_norm.weightF32[2560]
-
blk.26.post_ffw_norm.weightF32[2560]
-
blk.27.attn_k.weightF16[2560, 1024]
-
blk.27.attn_k_norm.weightF32[256]
-
blk.27.attn_norm.weightF32[2560]
-
blk.27.attn_output.weightF16[2048, 2560]
-
blk.27.attn_q.weightF16[2560, 2048]
-
blk.27.attn_q_norm.weightF32[256]
-
blk.27.attn_v.weightF16[2560, 1024]
-
blk.27.ffn_down.weightF16[10240, 2560]
-
blk.27.ffn_gate.weightF16[2560, 10240]
-
blk.27.ffn_norm.weightF32[2560]
-
blk.27.ffn_up.weightF16[2560, 10240]
-
blk.27.post_attention_norm.weightF32[2560]
-
blk.27.post_ffw_norm.weightF32[2560]
-
blk.28.attn_k.weightF16[2560, 1024]
-
blk.28.attn_k_norm.weightF32[256]
-
blk.28.attn_norm.weightF32[2560]
-
blk.28.attn_output.weightF16[2048, 2560]
-
blk.28.attn_q.weightF16[2560, 2048]
-
blk.28.attn_q_norm.weightF32[256]
-
blk.28.attn_v.weightF16[2560, 1024]
-
blk.28.ffn_down.weightF16[10240, 2560]
-
blk.28.ffn_gate.weightF16[2560, 10240]
-
blk.28.ffn_norm.weightF32[2560]
-
blk.28.ffn_up.weightF16[2560, 10240]
-
blk.28.post_attention_norm.weightF32[2560]
-
blk.28.post_ffw_norm.weightF32[2560]
-
blk.29.attn_k.weightF16[2560, 1024]
-
blk.29.attn_k_norm.weightF32[256]
-
blk.29.attn_norm.weightF32[2560]
-
blk.29.attn_output.weightF16[2048, 2560]
-
blk.29.attn_q.weightF16[2560, 2048]
-
blk.29.attn_q_norm.weightF32[256]
-
blk.29.attn_v.weightF16[2560, 1024]
-
blk.29.ffn_down.weightF16[10240, 2560]
-
blk.29.ffn_gate.weightF16[2560, 10240]
-
blk.29.ffn_norm.weightF32[2560]
-
blk.29.ffn_up.weightF16[2560, 10240]
-
blk.29.post_attention_norm.weightF32[2560]
-
blk.29.post_ffw_norm.weightF32[2560]
-
blk.30.attn_k.weightF16[2560, 1024]
-
blk.30.attn_k_norm.weightF32[256]
-
blk.30.attn_norm.weightF32[2560]
-
blk.30.attn_output.weightF16[2048, 2560]
-
blk.30.attn_q.weightF16[2560, 2048]
-
blk.30.attn_q_norm.weightF32[256]
-
blk.30.attn_v.weightF16[2560, 1024]
-
blk.30.ffn_down.weightF16[10240, 2560]
-
blk.30.ffn_gate.weightF16[2560, 10240]
-
blk.30.ffn_norm.weightF32[2560]
-
blk.30.ffn_up.weightF16[2560, 10240]
-
blk.30.post_attention_norm.weightF32[2560]
-
blk.30.post_ffw_norm.weightF32[2560]
-
blk.31.attn_k.weightF16[2560, 1024]
-
blk.31.attn_k_norm.weightF32[256]
-
blk.31.attn_norm.weightF32[2560]
-
blk.31.attn_output.weightF16[2048, 2560]
-
blk.31.attn_q.weightF16[2560, 2048]
-
blk.31.attn_q_norm.weightF32[256]
-
blk.31.attn_v.weightF16[2560, 1024]
-
blk.31.ffn_down.weightF16[10240, 2560]
-
blk.31.ffn_gate.weightF16[2560, 10240]
-
blk.31.ffn_norm.weightF32[2560]
-
blk.31.ffn_up.weightF16[2560, 10240]
-
blk.31.post_attention_norm.weightF32[2560]
-
blk.31.post_ffw_norm.weightF32[2560]
-
blk.32.attn_k.weightF16[2560, 1024]
-
blk.32.attn_k_norm.weightF32[256]
-
blk.32.attn_norm.weightF32[2560]
-
blk.32.attn_output.weightF16[2048, 2560]
-
blk.32.attn_q.weightF16[2560, 2048]
-
blk.32.attn_q_norm.weightF32[256]
-
blk.32.attn_v.weightF16[2560, 1024]
-
blk.32.ffn_down.weightF16[10240, 2560]
-
blk.32.ffn_gate.weightF16[2560, 10240]
-
blk.32.ffn_norm.weightF32[2560]
-
blk.32.ffn_up.weightF16[2560, 10240]
-
blk.32.post_attention_norm.weightF32[2560]
-
blk.32.post_ffw_norm.weightF32[2560]
-
blk.33.attn_k.weightF16[2560, 1024]
-
blk.33.attn_k_norm.weightF32[256]
-
blk.33.attn_norm.weightF32[2560]
-
blk.33.attn_output.weightF16[2048, 2560]
-
blk.33.attn_q.weightF16[2560, 2048]
-
blk.33.attn_q_norm.weightF32[256]
-
blk.33.attn_v.weightF16[2560, 1024]
-
blk.33.ffn_down.weightF16[10240, 2560]
-
blk.33.ffn_gate.weightF16[2560, 10240]
-
blk.33.ffn_norm.weightF32[2560]
-
blk.33.ffn_up.weightF16[2560, 10240]
-
blk.33.post_attention_norm.weightF32[2560]
-
blk.33.post_ffw_norm.weightF32[2560]
-
output_norm.weightF32[2560]
Metadata
Tensor
blk.0
blk.1
blk.2
blk.3
blk.4
blk.5
blk.6
blk.7
blk.8
blk.9
blk.10
blk.11
blk.12
blk.13
blk.14
blk.15
blk.16
blk.17
blk.18
blk.19
blk.20
blk.21
blk.22
blk.23
blk.24
blk.25
blk.26
blk.27
blk.28
blk.29
blk.30
blk.31
blk.32
blk.33