The IBM Granite 2B and 8B models are text-only dense LLMs trained on over 12 trillion tokens of data, demonstrated significant improvements over their predecessors in performance and speed in IBM’s initial testing.
tools
2b
8b
3,932 Pulls Updated 4 days ago
1f2aa59ffce4 · 1.5GB
-
general.architecturegranite
-
general.base_model.0.nameGranite 3.1 2b Base
-
general.base_model.0.organizationIbm Granite
-
general.base_model.0.repo_urlhttps://huggingface.co/ibm-granite/granite-3.1-2b-base
-
general.base_model.count1
-
general.basenamegranite-3.1
-
general.file_type2
-
general.finetuneinstruct
-
general.licenseapache-2.0
-
general.nameGranite 3.1 2b Instruct
-
general.quantization_version2
-
general.size_label2B
-
general.tags[language, granite-3.1, text-generation]
-
general.typemodel
-
granite.attention.head_count32
-
granite.attention.head_count_kv8
-
granite.attention.layer_norm_rms_epsilon1e-05
-
granite.attention.scale0.015625
-
granite.block_count40
-
granite.context_length131072
-
granite.embedding_length2048
-
granite.embedding_scale12
-
granite.feed_forward_length8192
-
granite.logit_scale8
-
granite.residual_scale0.22
-
granite.rope.dimension_count64
-
granite.rope.freq_base5e+06
-
granite.vocab_size49155
-
tokenizer.ggml.add_bos_tokenfalse
-
tokenizer.ggml.add_space_prefixfalse
-
tokenizer.ggml.bos_token_id0
-
tokenizer.ggml.eos_token_id0
-
tokenizer.ggml.merges[Ġ Ġ, ĠĠ ĠĠ, ĠĠĠĠ ĠĠĠĠ, ĠĠ Ġ, e r, ...]
-
tokenizer.ggml.modelgpt2
-
tokenizer.ggml.padding_token_id0
-
tokenizer.ggml.prerefact
-
tokenizer.ggml.token_type[3, 3, 3, 3, 3, ...]
-
tokenizer.ggml.tokens[<|end_of_text|>, <fim_prefix>, <fim_middle>, <fim_suffix>, <fim_pad>, ...]
-
tokenizer.ggml.unknown_token_id0
-
NameTypeShape
-
token_embd.weightQ6_K[2048, 49155]
-
blk.0.attn_k.weightQ4_0[2048, 512]
-
blk.0.attn_norm.weightF32[2048]
-
blk.0.attn_output.weightQ4_0[2048, 2048]
-
blk.0.attn_q.weightQ4_0[2048, 2048]
-
blk.0.attn_v.weightQ4_0[2048, 512]
-
blk.0.ffn_down.weightQ4_0[8192, 2048]
-
blk.0.ffn_gate.weightQ4_0[2048, 8192]
-
blk.0.ffn_norm.weightF32[2048]
-
blk.0.ffn_up.weightQ4_0[2048, 8192]
-
blk.1.attn_k.weightQ4_0[2048, 512]
-
blk.1.attn_norm.weightF32[2048]
-
blk.1.attn_output.weightQ4_0[2048, 2048]
-
blk.1.attn_q.weightQ4_0[2048, 2048]
-
blk.1.attn_v.weightQ4_0[2048, 512]
-
blk.1.ffn_down.weightQ4_0[8192, 2048]
-
blk.1.ffn_gate.weightQ4_0[2048, 8192]
-
blk.1.ffn_norm.weightF32[2048]
-
blk.1.ffn_up.weightQ4_0[2048, 8192]
-
blk.2.attn_k.weightQ4_0[2048, 512]
-
blk.2.attn_norm.weightF32[2048]
-
blk.2.attn_output.weightQ4_0[2048, 2048]
-
blk.2.attn_q.weightQ4_0[2048, 2048]
-
blk.2.attn_v.weightQ4_0[2048, 512]
-
blk.2.ffn_down.weightQ4_0[8192, 2048]
-
blk.2.ffn_gate.weightQ4_0[2048, 8192]
-
blk.2.ffn_norm.weightF32[2048]
-
blk.2.ffn_up.weightQ4_0[2048, 8192]
-
blk.3.attn_k.weightQ4_0[2048, 512]
-
blk.3.attn_norm.weightF32[2048]
-
blk.3.attn_output.weightQ4_0[2048, 2048]
-
blk.3.attn_q.weightQ4_0[2048, 2048]
-
blk.3.attn_v.weightQ4_0[2048, 512]
-
blk.3.ffn_down.weightQ4_0[8192, 2048]
-
blk.3.ffn_gate.weightQ4_0[2048, 8192]
-
blk.3.ffn_norm.weightF32[2048]
-
blk.3.ffn_up.weightQ4_0[2048, 8192]
-
blk.4.attn_k.weightQ4_0[2048, 512]
-
blk.4.attn_norm.weightF32[2048]
-
blk.4.attn_output.weightQ4_0[2048, 2048]
-
blk.4.attn_q.weightQ4_0[2048, 2048]
-
blk.4.attn_v.weightQ4_0[2048, 512]
-
blk.4.ffn_down.weightQ4_0[8192, 2048]
-
blk.4.ffn_gate.weightQ4_0[2048, 8192]
-
blk.4.ffn_norm.weightF32[2048]
-
blk.4.ffn_up.weightQ4_0[2048, 8192]
-
blk.5.attn_k.weightQ4_0[2048, 512]
-
blk.5.attn_norm.weightF32[2048]
-
blk.5.attn_output.weightQ4_0[2048, 2048]
-
blk.5.attn_q.weightQ4_0[2048, 2048]
-
blk.5.attn_v.weightQ4_0[2048, 512]
-
blk.5.ffn_down.weightQ4_0[8192, 2048]
-
blk.5.ffn_gate.weightQ4_0[2048, 8192]
-
blk.5.ffn_norm.weightF32[2048]
-
blk.5.ffn_up.weightQ4_0[2048, 8192]
-
blk.6.attn_k.weightQ4_0[2048, 512]
-
blk.6.attn_norm.weightF32[2048]
-
blk.6.attn_output.weightQ4_0[2048, 2048]
-
blk.6.attn_q.weightQ4_0[2048, 2048]
-
blk.6.attn_v.weightQ4_0[2048, 512]
-
blk.6.ffn_down.weightQ4_0[8192, 2048]
-
blk.6.ffn_gate.weightQ4_0[2048, 8192]
-
blk.6.ffn_norm.weightF32[2048]
-
blk.6.ffn_up.weightQ4_0[2048, 8192]
-
blk.7.attn_k.weightQ4_0[2048, 512]
-
blk.7.attn_norm.weightF32[2048]
-
blk.7.attn_output.weightQ4_0[2048, 2048]
-
blk.7.attn_q.weightQ4_0[2048, 2048]
-
blk.7.attn_v.weightQ4_0[2048, 512]
-
blk.7.ffn_down.weightQ4_0[8192, 2048]
-
blk.7.ffn_gate.weightQ4_0[2048, 8192]
-
blk.7.ffn_norm.weightF32[2048]
-
blk.7.ffn_up.weightQ4_0[2048, 8192]
-
blk.8.attn_k.weightQ4_0[2048, 512]
-
blk.8.attn_norm.weightF32[2048]
-
blk.8.attn_output.weightQ4_0[2048, 2048]
-
blk.8.attn_q.weightQ4_0[2048, 2048]
-
blk.8.attn_v.weightQ4_0[2048, 512]
-
blk.8.ffn_down.weightQ4_0[8192, 2048]
-
blk.8.ffn_gate.weightQ4_0[2048, 8192]
-
blk.8.ffn_norm.weightF32[2048]
-
blk.8.ffn_up.weightQ4_0[2048, 8192]
-
blk.9.attn_k.weightQ4_0[2048, 512]
-
blk.9.attn_norm.weightF32[2048]
-
blk.9.attn_output.weightQ4_0[2048, 2048]
-
blk.9.attn_q.weightQ4_0[2048, 2048]
-
blk.9.attn_v.weightQ4_0[2048, 512]
-
blk.9.ffn_down.weightQ4_0[8192, 2048]
-
blk.9.ffn_gate.weightQ4_0[2048, 8192]
-
blk.9.ffn_norm.weightF32[2048]
-
blk.9.ffn_up.weightQ4_0[2048, 8192]
-
blk.10.attn_k.weightQ4_0[2048, 512]
-
blk.10.attn_norm.weightF32[2048]
-
blk.10.attn_output.weightQ4_0[2048, 2048]
-
blk.10.attn_q.weightQ4_0[2048, 2048]
-
blk.10.attn_v.weightQ4_0[2048, 512]
-
blk.10.ffn_down.weightQ4_0[8192, 2048]
-
blk.10.ffn_gate.weightQ4_0[2048, 8192]
-
blk.10.ffn_norm.weightF32[2048]
-
blk.10.ffn_up.weightQ4_0[2048, 8192]
-
blk.11.attn_k.weightQ4_0[2048, 512]
-
blk.11.attn_norm.weightF32[2048]
-
blk.11.attn_output.weightQ4_0[2048, 2048]
-
blk.11.attn_q.weightQ4_0[2048, 2048]
-
blk.11.attn_v.weightQ4_0[2048, 512]
-
blk.11.ffn_down.weightQ4_0[8192, 2048]
-
blk.11.ffn_gate.weightQ4_0[2048, 8192]
-
blk.11.ffn_norm.weightF32[2048]
-
blk.11.ffn_up.weightQ4_0[2048, 8192]
-
blk.12.attn_k.weightQ4_0[2048, 512]
-
blk.12.attn_norm.weightF32[2048]
-
blk.12.attn_output.weightQ4_0[2048, 2048]
-
blk.12.attn_q.weightQ4_0[2048, 2048]
-
blk.12.attn_v.weightQ4_0[2048, 512]
-
blk.12.ffn_down.weightQ4_0[8192, 2048]
-
blk.12.ffn_gate.weightQ4_0[2048, 8192]
-
blk.12.ffn_norm.weightF32[2048]
-
blk.12.ffn_up.weightQ4_0[2048, 8192]
-
blk.13.attn_k.weightQ4_0[2048, 512]
-
blk.13.attn_norm.weightF32[2048]
-
blk.13.attn_output.weightQ4_0[2048, 2048]
-
blk.13.attn_q.weightQ4_0[2048, 2048]
-
blk.13.attn_v.weightQ4_0[2048, 512]
-
blk.13.ffn_down.weightQ4_0[8192, 2048]
-
blk.13.ffn_gate.weightQ4_0[2048, 8192]
-
blk.13.ffn_norm.weightF32[2048]
-
blk.13.ffn_up.weightQ4_0[2048, 8192]
-
blk.14.attn_k.weightQ4_0[2048, 512]
-
blk.14.attn_norm.weightF32[2048]
-
blk.14.attn_output.weightQ4_0[2048, 2048]
-
blk.14.attn_q.weightQ4_0[2048, 2048]
-
blk.14.attn_v.weightQ4_0[2048, 512]
-
blk.14.ffn_down.weightQ4_0[8192, 2048]
-
blk.14.ffn_gate.weightQ4_0[2048, 8192]
-
blk.14.ffn_norm.weightF32[2048]
-
blk.14.ffn_up.weightQ4_0[2048, 8192]
-
blk.15.attn_k.weightQ4_0[2048, 512]
-
blk.15.attn_norm.weightF32[2048]
-
blk.15.attn_output.weightQ4_0[2048, 2048]
-
blk.15.attn_q.weightQ4_0[2048, 2048]
-
blk.15.attn_v.weightQ4_0[2048, 512]
-
blk.15.ffn_down.weightQ4_0[8192, 2048]
-
blk.15.ffn_gate.weightQ4_0[2048, 8192]
-
blk.15.ffn_norm.weightF32[2048]
-
blk.15.ffn_up.weightQ4_0[2048, 8192]
-
blk.16.attn_k.weightQ4_0[2048, 512]
-
blk.16.attn_norm.weightF32[2048]
-
blk.16.attn_output.weightQ4_0[2048, 2048]
-
blk.16.attn_q.weightQ4_0[2048, 2048]
-
blk.16.attn_v.weightQ4_0[2048, 512]
-
blk.16.ffn_down.weightQ4_0[8192, 2048]
-
blk.16.ffn_gate.weightQ4_0[2048, 8192]
-
blk.16.ffn_norm.weightF32[2048]
-
blk.16.ffn_up.weightQ4_0[2048, 8192]
-
blk.17.attn_k.weightQ4_0[2048, 512]
-
blk.17.attn_norm.weightF32[2048]
-
blk.17.attn_output.weightQ4_0[2048, 2048]
-
blk.17.attn_q.weightQ4_0[2048, 2048]
-
blk.17.attn_v.weightQ4_0[2048, 512]
-
blk.17.ffn_down.weightQ4_0[8192, 2048]
-
blk.17.ffn_gate.weightQ4_0[2048, 8192]
-
blk.17.ffn_norm.weightF32[2048]
-
blk.17.ffn_up.weightQ4_0[2048, 8192]
-
blk.18.attn_k.weightQ4_0[2048, 512]
-
blk.18.attn_norm.weightF32[2048]
-
blk.18.attn_output.weightQ4_0[2048, 2048]
-
blk.18.attn_q.weightQ4_0[2048, 2048]
-
blk.18.attn_v.weightQ4_0[2048, 512]
-
blk.18.ffn_down.weightQ4_0[8192, 2048]
-
blk.18.ffn_gate.weightQ4_0[2048, 8192]
-
blk.18.ffn_norm.weightF32[2048]
-
blk.18.ffn_up.weightQ4_0[2048, 8192]
-
blk.19.attn_k.weightQ4_0[2048, 512]
-
blk.19.attn_norm.weightF32[2048]
-
blk.19.attn_output.weightQ4_0[2048, 2048]
-
blk.19.attn_q.weightQ4_0[2048, 2048]
-
blk.19.attn_v.weightQ4_0[2048, 512]
-
blk.19.ffn_down.weightQ4_0[8192, 2048]
-
blk.19.ffn_gate.weightQ4_0[2048, 8192]
-
blk.19.ffn_norm.weightF32[2048]
-
blk.19.ffn_up.weightQ4_0[2048, 8192]
-
blk.20.attn_k.weightQ4_0[2048, 512]
-
blk.20.attn_norm.weightF32[2048]
-
blk.20.attn_output.weightQ4_0[2048, 2048]
-
blk.20.attn_q.weightQ4_0[2048, 2048]
-
blk.20.attn_v.weightQ4_0[2048, 512]
-
blk.20.ffn_down.weightQ4_0[8192, 2048]
-
blk.20.ffn_gate.weightQ4_0[2048, 8192]
-
blk.20.ffn_norm.weightF32[2048]
-
blk.20.ffn_up.weightQ4_0[2048, 8192]
-
blk.21.attn_k.weightQ4_0[2048, 512]
-
blk.21.attn_norm.weightF32[2048]
-
blk.21.attn_output.weightQ4_0[2048, 2048]
-
blk.21.attn_q.weightQ4_0[2048, 2048]
-
blk.21.attn_v.weightQ4_0[2048, 512]
-
blk.21.ffn_down.weightQ4_0[8192, 2048]
-
blk.21.ffn_gate.weightQ4_0[2048, 8192]
-
blk.21.ffn_norm.weightF32[2048]
-
blk.21.ffn_up.weightQ4_0[2048, 8192]
-
blk.22.attn_k.weightQ4_0[2048, 512]
-
blk.22.attn_norm.weightF32[2048]
-
blk.22.attn_output.weightQ4_0[2048, 2048]
-
blk.22.attn_q.weightQ4_0[2048, 2048]
-
blk.22.attn_v.weightQ4_0[2048, 512]
-
blk.22.ffn_down.weightQ4_0[8192, 2048]
-
blk.22.ffn_gate.weightQ4_0[2048, 8192]
-
blk.22.ffn_norm.weightF32[2048]
-
blk.22.ffn_up.weightQ4_0[2048, 8192]
-
blk.23.attn_k.weightQ4_0[2048, 512]
-
blk.23.attn_norm.weightF32[2048]
-
blk.23.attn_output.weightQ4_0[2048, 2048]
-
blk.23.attn_q.weightQ4_0[2048, 2048]
-
blk.23.attn_v.weightQ4_0[2048, 512]
-
blk.23.ffn_down.weightQ4_0[8192, 2048]
-
blk.23.ffn_gate.weightQ4_0[2048, 8192]
-
blk.23.ffn_norm.weightF32[2048]
-
blk.23.ffn_up.weightQ4_0[2048, 8192]
-
blk.24.attn_k.weightQ4_0[2048, 512]
-
blk.24.attn_norm.weightF32[2048]
-
blk.24.attn_output.weightQ4_0[2048, 2048]
-
blk.24.attn_q.weightQ4_0[2048, 2048]
-
blk.24.attn_v.weightQ4_0[2048, 512]
-
blk.24.ffn_down.weightQ4_0[8192, 2048]
-
blk.24.ffn_gate.weightQ4_0[2048, 8192]
-
blk.24.ffn_norm.weightF32[2048]
-
blk.24.ffn_up.weightQ4_0[2048, 8192]
-
blk.25.attn_k.weightQ4_0[2048, 512]
-
blk.25.attn_norm.weightF32[2048]
-
blk.25.attn_output.weightQ4_0[2048, 2048]
-
blk.25.attn_q.weightQ4_0[2048, 2048]
-
blk.25.attn_v.weightQ4_0[2048, 512]
-
blk.25.ffn_down.weightQ4_0[8192, 2048]
-
blk.25.ffn_gate.weightQ4_0[2048, 8192]
-
blk.25.ffn_norm.weightF32[2048]
-
blk.25.ffn_up.weightQ4_0[2048, 8192]
-
blk.26.attn_k.weightQ4_0[2048, 512]
-
blk.26.attn_norm.weightF32[2048]
-
blk.26.attn_output.weightQ4_0[2048, 2048]
-
blk.26.attn_q.weightQ4_0[2048, 2048]
-
blk.26.attn_v.weightQ4_0[2048, 512]
-
blk.26.ffn_down.weightQ4_0[8192, 2048]
-
blk.26.ffn_gate.weightQ4_0[2048, 8192]
-
blk.26.ffn_norm.weightF32[2048]
-
blk.26.ffn_up.weightQ4_0[2048, 8192]
-
blk.27.attn_k.weightQ4_0[2048, 512]
-
blk.27.attn_norm.weightF32[2048]
-
blk.27.attn_output.weightQ4_0[2048, 2048]
-
blk.27.attn_q.weightQ4_0[2048, 2048]
-
blk.27.attn_v.weightQ4_0[2048, 512]
-
blk.27.ffn_down.weightQ4_0[8192, 2048]
-
blk.27.ffn_gate.weightQ4_0[2048, 8192]
-
blk.27.ffn_norm.weightF32[2048]
-
blk.27.ffn_up.weightQ4_0[2048, 8192]
-
blk.28.attn_k.weightQ4_0[2048, 512]
-
blk.28.attn_norm.weightF32[2048]
-
blk.28.attn_output.weightQ4_0[2048, 2048]
-
blk.28.attn_q.weightQ4_0[2048, 2048]
-
blk.28.attn_v.weightQ4_0[2048, 512]
-
blk.28.ffn_down.weightQ4_0[8192, 2048]
-
blk.28.ffn_gate.weightQ4_0[2048, 8192]
-
blk.28.ffn_norm.weightF32[2048]
-
blk.28.ffn_up.weightQ4_0[2048, 8192]
-
blk.29.attn_k.weightQ4_0[2048, 512]
-
blk.29.attn_norm.weightF32[2048]
-
blk.29.attn_output.weightQ4_0[2048, 2048]
-
blk.29.attn_q.weightQ4_0[2048, 2048]
-
blk.29.attn_v.weightQ4_0[2048, 512]
-
blk.29.ffn_down.weightQ4_0[8192, 2048]
-
blk.29.ffn_gate.weightQ4_0[2048, 8192]
-
blk.29.ffn_norm.weightF32[2048]
-
blk.29.ffn_up.weightQ4_0[2048, 8192]
-
blk.30.attn_k.weightQ4_0[2048, 512]
-
blk.30.attn_norm.weightF32[2048]
-
blk.30.attn_output.weightQ4_0[2048, 2048]
-
blk.30.attn_q.weightQ4_0[2048, 2048]
-
blk.30.attn_v.weightQ4_0[2048, 512]
-
blk.30.ffn_down.weightQ4_0[8192, 2048]
-
blk.30.ffn_gate.weightQ4_0[2048, 8192]
-
blk.30.ffn_norm.weightF32[2048]
-
blk.30.ffn_up.weightQ4_0[2048, 8192]
-
blk.31.attn_k.weightQ4_0[2048, 512]
-
blk.31.attn_norm.weightF32[2048]
-
blk.31.attn_output.weightQ4_0[2048, 2048]
-
blk.31.attn_q.weightQ4_0[2048, 2048]
-
blk.31.attn_v.weightQ4_0[2048, 512]
-
blk.31.ffn_down.weightQ4_0[8192, 2048]
-
blk.31.ffn_gate.weightQ4_0[2048, 8192]
-
blk.31.ffn_norm.weightF32[2048]
-
blk.31.ffn_up.weightQ4_0[2048, 8192]
-
blk.32.attn_k.weightQ4_0[2048, 512]
-
blk.32.attn_norm.weightF32[2048]
-
blk.32.attn_output.weightQ4_0[2048, 2048]
-
blk.32.attn_q.weightQ4_0[2048, 2048]
-
blk.32.attn_v.weightQ4_0[2048, 512]
-
blk.32.ffn_down.weightQ4_0[8192, 2048]
-
blk.32.ffn_gate.weightQ4_0[2048, 8192]
-
blk.32.ffn_norm.weightF32[2048]
-
blk.32.ffn_up.weightQ4_0[2048, 8192]
-
blk.33.attn_k.weightQ4_0[2048, 512]
-
blk.33.attn_norm.weightF32[2048]
-
blk.33.attn_output.weightQ4_0[2048, 2048]
-
blk.33.attn_q.weightQ4_0[2048, 2048]
-
blk.33.attn_v.weightQ4_0[2048, 512]
-
blk.33.ffn_down.weightQ4_0[8192, 2048]
-
blk.33.ffn_gate.weightQ4_0[2048, 8192]
-
blk.33.ffn_norm.weightF32[2048]
-
blk.33.ffn_up.weightQ4_0[2048, 8192]
-
blk.34.attn_k.weightQ4_0[2048, 512]
-
blk.34.attn_norm.weightF32[2048]
-
blk.34.attn_output.weightQ4_0[2048, 2048]
-
blk.34.attn_q.weightQ4_0[2048, 2048]
-
blk.34.attn_v.weightQ4_0[2048, 512]
-
blk.34.ffn_down.weightQ4_0[8192, 2048]
-
blk.34.ffn_gate.weightQ4_0[2048, 8192]
-
blk.34.ffn_norm.weightF32[2048]
-
blk.34.ffn_up.weightQ4_0[2048, 8192]
-
blk.35.attn_k.weightQ4_0[2048, 512]
-
blk.35.attn_norm.weightF32[2048]
-
blk.35.attn_output.weightQ4_0[2048, 2048]
-
blk.35.attn_q.weightQ4_0[2048, 2048]
-
blk.35.attn_v.weightQ4_0[2048, 512]
-
blk.35.ffn_down.weightQ4_0[8192, 2048]
-
blk.35.ffn_gate.weightQ4_0[2048, 8192]
-
blk.35.ffn_norm.weightF32[2048]
-
blk.35.ffn_up.weightQ4_0[2048, 8192]
-
blk.36.attn_k.weightQ4_0[2048, 512]
-
blk.36.attn_norm.weightF32[2048]
-
blk.36.attn_output.weightQ4_0[2048, 2048]
-
blk.36.attn_q.weightQ4_0[2048, 2048]
-
blk.36.attn_v.weightQ4_0[2048, 512]
-
blk.36.ffn_down.weightQ4_0[8192, 2048]
-
blk.36.ffn_gate.weightQ4_0[2048, 8192]
-
blk.36.ffn_norm.weightF32[2048]
-
blk.36.ffn_up.weightQ4_0[2048, 8192]
-
blk.37.attn_k.weightQ4_0[2048, 512]
-
blk.37.attn_norm.weightF32[2048]
-
blk.37.attn_output.weightQ4_0[2048, 2048]
-
blk.37.attn_q.weightQ4_0[2048, 2048]
-
blk.37.attn_v.weightQ4_0[2048, 512]
-
blk.37.ffn_down.weightQ4_0[8192, 2048]
-
blk.37.ffn_gate.weightQ4_0[2048, 8192]
-
blk.37.ffn_norm.weightF32[2048]
-
blk.37.ffn_up.weightQ4_0[2048, 8192]
-
blk.38.attn_k.weightQ4_0[2048, 512]
-
blk.38.attn_norm.weightF32[2048]
-
blk.38.attn_output.weightQ4_0[2048, 2048]
-
blk.38.attn_q.weightQ4_0[2048, 2048]
-
blk.38.attn_v.weightQ4_0[2048, 512]
-
blk.38.ffn_down.weightQ4_0[8192, 2048]
-
blk.38.ffn_gate.weightQ4_0[2048, 8192]
-
blk.38.ffn_norm.weightF32[2048]
-
blk.38.ffn_up.weightQ4_0[2048, 8192]
-
blk.39.attn_k.weightQ4_0[2048, 512]
-
blk.39.attn_norm.weightF32[2048]
-
blk.39.attn_output.weightQ4_0[2048, 2048]
-
blk.39.attn_q.weightQ4_0[2048, 2048]
-
blk.39.attn_v.weightQ4_0[2048, 512]
-
blk.39.ffn_down.weightQ4_0[8192, 2048]
-
blk.39.ffn_gate.weightQ4_0[2048, 8192]
-
blk.39.ffn_norm.weightF32[2048]
-
blk.39.ffn_up.weightQ4_0[2048, 8192]
-
output_norm.weightF32[2048]
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
blk.34
blk.35
blk.36
blk.37
blk.38
blk.39