H2O Danube3 (0.5B & 4B)
0.5b
4b
182 Pulls Updated 4 months ago
a9ceb91e7e83 · 547MB
-
general.architecturellama
-
general.file_type7
-
general.namedanube3-500M
-
general.quantization_version2
-
llama.attention.head_count16
-
llama.attention.head_count_kv8
-
llama.attention.layer_norm_rms_epsilon1e-05
-
llama.block_count16
-
llama.context_length8192
-
llama.embedding_length1536
-
llama.feed_forward_length4096
-
llama.rope.dimension_count96
-
llama.rope.freq_base100000
-
llama.vocab_size32000
-
tokenizer.ggml.add_bos_tokenfalse
-
tokenizer.ggml.add_eos_tokenfalse
-
tokenizer.ggml.bos_token_id1
-
tokenizer.ggml.cls_token_id2
-
tokenizer.ggml.eos_token_id2
-
tokenizer.ggml.modelllama
-
tokenizer.ggml.padding_token_id0
-
tokenizer.ggml.predefault
-
tokenizer.ggml.scores[0, 0, 0, 0, 0, ...]
-
tokenizer.ggml.seperator_token_id2
-
tokenizer.ggml.token_type[2, 3, 3, 6, 6, ...]
-
tokenizer.ggml.tokens[<unk>, <s>, </s>, <0x00>, <0x01>, ...]
-
tokenizer.ggml.unknown_token_id0
-
NameTypeShape
-
token_embd.weightQ8_0[1536, 32000]
-
blk.0.attn_k.weightQ8_0[1536, 768]
-
blk.0.attn_norm.weightF32[1536]
-
blk.0.attn_output.weightQ8_0[1536, 1536]
-
blk.0.attn_q.weightQ8_0[1536, 1536]
-
blk.0.attn_v.weightQ8_0[1536, 768]
-
blk.0.ffn_down.weightQ8_0[4096, 1536]
-
blk.0.ffn_gate.weightQ8_0[1536, 4096]
-
blk.0.ffn_norm.weightF32[1536]
-
blk.0.ffn_up.weightQ8_0[1536, 4096]
-
blk.1.attn_k.weightQ8_0[1536, 768]
-
blk.1.attn_norm.weightF32[1536]
-
blk.1.attn_output.weightQ8_0[1536, 1536]
-
blk.1.attn_q.weightQ8_0[1536, 1536]
-
blk.1.attn_v.weightQ8_0[1536, 768]
-
blk.1.ffn_down.weightQ8_0[4096, 1536]
-
blk.1.ffn_gate.weightQ8_0[1536, 4096]
-
blk.1.ffn_norm.weightF32[1536]
-
blk.1.ffn_up.weightQ8_0[1536, 4096]
-
blk.2.attn_k.weightQ8_0[1536, 768]
-
blk.2.attn_norm.weightF32[1536]
-
blk.2.attn_output.weightQ8_0[1536, 1536]
-
blk.2.attn_q.weightQ8_0[1536, 1536]
-
blk.2.attn_v.weightQ8_0[1536, 768]
-
blk.2.ffn_down.weightQ8_0[4096, 1536]
-
blk.2.ffn_gate.weightQ8_0[1536, 4096]
-
blk.2.ffn_norm.weightF32[1536]
-
blk.2.ffn_up.weightQ8_0[1536, 4096]
-
blk.3.attn_k.weightQ8_0[1536, 768]
-
blk.3.attn_norm.weightF32[1536]
-
blk.3.attn_output.weightQ8_0[1536, 1536]
-
blk.3.attn_q.weightQ8_0[1536, 1536]
-
blk.3.attn_v.weightQ8_0[1536, 768]
-
blk.3.ffn_down.weightQ8_0[4096, 1536]
-
blk.3.ffn_gate.weightQ8_0[1536, 4096]
-
blk.3.ffn_norm.weightF32[1536]
-
blk.3.ffn_up.weightQ8_0[1536, 4096]
-
blk.4.attn_k.weightQ8_0[1536, 768]
-
blk.4.attn_norm.weightF32[1536]
-
blk.4.attn_output.weightQ8_0[1536, 1536]
-
blk.4.attn_q.weightQ8_0[1536, 1536]
-
blk.4.attn_v.weightQ8_0[1536, 768]
-
blk.4.ffn_down.weightQ8_0[4096, 1536]
-
blk.4.ffn_gate.weightQ8_0[1536, 4096]
-
blk.4.ffn_norm.weightF32[1536]
-
blk.4.ffn_up.weightQ8_0[1536, 4096]
-
blk.5.attn_k.weightQ8_0[1536, 768]
-
blk.5.attn_norm.weightF32[1536]
-
blk.5.attn_output.weightQ8_0[1536, 1536]
-
blk.5.attn_q.weightQ8_0[1536, 1536]
-
blk.5.attn_v.weightQ8_0[1536, 768]
-
blk.5.ffn_down.weightQ8_0[4096, 1536]
-
blk.5.ffn_gate.weightQ8_0[1536, 4096]
-
blk.5.ffn_norm.weightF32[1536]
-
blk.5.ffn_up.weightQ8_0[1536, 4096]
-
blk.6.attn_k.weightQ8_0[1536, 768]
-
blk.6.attn_norm.weightF32[1536]
-
blk.6.attn_output.weightQ8_0[1536, 1536]
-
blk.6.attn_q.weightQ8_0[1536, 1536]
-
blk.6.attn_v.weightQ8_0[1536, 768]
-
blk.6.ffn_down.weightQ8_0[4096, 1536]
-
blk.6.ffn_gate.weightQ8_0[1536, 4096]
-
blk.6.ffn_norm.weightF32[1536]
-
blk.6.ffn_up.weightQ8_0[1536, 4096]
-
blk.7.attn_k.weightQ8_0[1536, 768]
-
blk.7.attn_norm.weightF32[1536]
-
blk.7.attn_output.weightQ8_0[1536, 1536]
-
blk.7.attn_q.weightQ8_0[1536, 1536]
-
blk.7.attn_v.weightQ8_0[1536, 768]
-
blk.7.ffn_down.weightQ8_0[4096, 1536]
-
blk.7.ffn_gate.weightQ8_0[1536, 4096]
-
blk.7.ffn_norm.weightF32[1536]
-
blk.7.ffn_up.weightQ8_0[1536, 4096]
-
blk.8.attn_k.weightQ8_0[1536, 768]
-
blk.8.attn_norm.weightF32[1536]
-
blk.8.attn_output.weightQ8_0[1536, 1536]
-
blk.8.attn_q.weightQ8_0[1536, 1536]
-
blk.8.attn_v.weightQ8_0[1536, 768]
-
blk.8.ffn_down.weightQ8_0[4096, 1536]
-
blk.8.ffn_gate.weightQ8_0[1536, 4096]
-
blk.8.ffn_norm.weightF32[1536]
-
blk.8.ffn_up.weightQ8_0[1536, 4096]
-
blk.9.attn_k.weightQ8_0[1536, 768]
-
blk.9.attn_norm.weightF32[1536]
-
blk.9.attn_output.weightQ8_0[1536, 1536]
-
blk.9.attn_q.weightQ8_0[1536, 1536]
-
blk.9.attn_v.weightQ8_0[1536, 768]
-
blk.9.ffn_down.weightQ8_0[4096, 1536]
-
blk.9.ffn_gate.weightQ8_0[1536, 4096]
-
blk.9.ffn_norm.weightF32[1536]
-
blk.9.ffn_up.weightQ8_0[1536, 4096]
-
blk.10.attn_k.weightQ8_0[1536, 768]
-
blk.10.attn_norm.weightF32[1536]
-
blk.10.attn_output.weightQ8_0[1536, 1536]
-
blk.10.attn_q.weightQ8_0[1536, 1536]
-
blk.10.attn_v.weightQ8_0[1536, 768]
-
blk.10.ffn_down.weightQ8_0[4096, 1536]
-
blk.10.ffn_gate.weightQ8_0[1536, 4096]
-
blk.10.ffn_norm.weightF32[1536]
-
blk.10.ffn_up.weightQ8_0[1536, 4096]
-
blk.11.attn_k.weightQ8_0[1536, 768]
-
blk.11.attn_norm.weightF32[1536]
-
blk.11.attn_output.weightQ8_0[1536, 1536]
-
blk.11.attn_q.weightQ8_0[1536, 1536]
-
blk.11.attn_v.weightQ8_0[1536, 768]
-
blk.11.ffn_down.weightQ8_0[4096, 1536]
-
blk.11.ffn_gate.weightQ8_0[1536, 4096]
-
blk.11.ffn_norm.weightF32[1536]
-
blk.11.ffn_up.weightQ8_0[1536, 4096]
-
blk.12.attn_k.weightQ8_0[1536, 768]
-
blk.12.attn_norm.weightF32[1536]
-
blk.12.attn_output.weightQ8_0[1536, 1536]
-
blk.12.attn_q.weightQ8_0[1536, 1536]
-
blk.12.attn_v.weightQ8_0[1536, 768]
-
blk.12.ffn_down.weightQ8_0[4096, 1536]
-
blk.12.ffn_gate.weightQ8_0[1536, 4096]
-
blk.12.ffn_norm.weightF32[1536]
-
blk.12.ffn_up.weightQ8_0[1536, 4096]
-
blk.13.attn_k.weightQ8_0[1536, 768]
-
blk.13.attn_norm.weightF32[1536]
-
blk.13.attn_output.weightQ8_0[1536, 1536]
-
blk.13.attn_q.weightQ8_0[1536, 1536]
-
blk.13.attn_v.weightQ8_0[1536, 768]
-
blk.13.ffn_down.weightQ8_0[4096, 1536]
-
blk.13.ffn_gate.weightQ8_0[1536, 4096]
-
blk.13.ffn_norm.weightF32[1536]
-
blk.13.ffn_up.weightQ8_0[1536, 4096]
-
blk.14.attn_k.weightQ8_0[1536, 768]
-
blk.14.attn_norm.weightF32[1536]
-
blk.14.attn_output.weightQ8_0[1536, 1536]
-
blk.14.attn_q.weightQ8_0[1536, 1536]
-
blk.14.attn_v.weightQ8_0[1536, 768]
-
blk.14.ffn_down.weightQ8_0[4096, 1536]
-
blk.14.ffn_gate.weightQ8_0[1536, 4096]
-
blk.14.ffn_norm.weightF32[1536]
-
blk.14.ffn_up.weightQ8_0[1536, 4096]
-
blk.15.attn_k.weightQ8_0[1536, 768]
-
blk.15.attn_norm.weightF32[1536]
-
blk.15.attn_output.weightQ8_0[1536, 1536]
-
blk.15.attn_q.weightQ8_0[1536, 1536]
-
blk.15.attn_v.weightQ8_0[1536, 768]
-
blk.15.ffn_down.weightQ8_0[4096, 1536]
-
blk.15.ffn_gate.weightQ8_0[1536, 4096]
-
blk.15.ffn_norm.weightF32[1536]
-
blk.15.ffn_up.weightQ8_0[1536, 4096]
-
output.weightQ8_0[1536, 32000]
-
output_norm.weightF32[1536]
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