The IBM Granite 1B and 3B models are long-context mixture of experts (MoE) Granite models from IBM designed for low latency usage.
tools
1b
3b
1,537 Pulls Updated 3 days ago
cd60b3e8bb44 · 1.4GB
-
general.architecturegranitemoe
-
general.base_model.0.nameGranite 3.1 1b A400M Base
-
general.base_model.0.organizationIbm Granite
-
general.base_model.0.repo_urlhttps://huggingface.co/ibm-granite/granite-3.1-1b-a400m-base
-
general.base_model.count1
-
general.basenamegranite-3.1
-
general.file_type7
-
general.finetuneinstruct
-
general.licenseapache-2.0
-
general.nameGranite 3.1 1b A400M Instruct
-
general.quantization_version2
-
general.size_label1B-a400M
-
general.tags[language, granite-3.1, text-generation]
-
general.typemodel
-
granitemoe.attention.head_count16
-
granitemoe.attention.head_count_kv8
-
granitemoe.attention.layer_norm_rms_epsilon1e-06
-
granitemoe.attention.scale0.015625
-
granitemoe.block_count24
-
granitemoe.context_length131072
-
granitemoe.embedding_length1024
-
granitemoe.embedding_scale12
-
granitemoe.expert_count32
-
granitemoe.expert_used_count8
-
granitemoe.feed_forward_length512
-
granitemoe.logit_scale6
-
granitemoe.residual_scale0.22
-
granitemoe.rope.dimension_count64
-
granitemoe.rope.freq_base1.5e+06
-
granitemoe.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.weightQ8_0[1024, 49155]
-
blk.0.attn_k.weightQ8_0[1024, 512]
-
blk.0.attn_norm.weightF32[1024]
-
blk.0.attn_output.weightQ8_0[1024, 1024]
-
blk.0.attn_q.weightQ8_0[1024, 1024]
-
blk.0.attn_v.weightQ8_0[1024, 512]
-
blk.0.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.0.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.0.ffn_gate_inp.weightF32[1024, 32]
-
blk.0.ffn_norm.weightF32[1024]
-
blk.0.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.1.attn_k.weightQ8_0[1024, 512]
-
blk.1.attn_norm.weightF32[1024]
-
blk.1.attn_output.weightQ8_0[1024, 1024]
-
blk.1.attn_q.weightQ8_0[1024, 1024]
-
blk.1.attn_v.weightQ8_0[1024, 512]
-
blk.1.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.1.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.1.ffn_gate_inp.weightF32[1024, 32]
-
blk.1.ffn_norm.weightF32[1024]
-
blk.1.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.2.attn_k.weightQ8_0[1024, 512]
-
blk.2.attn_norm.weightF32[1024]
-
blk.2.attn_output.weightQ8_0[1024, 1024]
-
blk.2.attn_q.weightQ8_0[1024, 1024]
-
blk.2.attn_v.weightQ8_0[1024, 512]
-
blk.2.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.2.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.2.ffn_gate_inp.weightF32[1024, 32]
-
blk.2.ffn_norm.weightF32[1024]
-
blk.2.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.3.attn_k.weightQ8_0[1024, 512]
-
blk.3.attn_norm.weightF32[1024]
-
blk.3.attn_output.weightQ8_0[1024, 1024]
-
blk.3.attn_q.weightQ8_0[1024, 1024]
-
blk.3.attn_v.weightQ8_0[1024, 512]
-
blk.3.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.3.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.3.ffn_gate_inp.weightF32[1024, 32]
-
blk.3.ffn_norm.weightF32[1024]
-
blk.3.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.4.attn_k.weightQ8_0[1024, 512]
-
blk.4.attn_norm.weightF32[1024]
-
blk.4.attn_output.weightQ8_0[1024, 1024]
-
blk.4.attn_q.weightQ8_0[1024, 1024]
-
blk.4.attn_v.weightQ8_0[1024, 512]
-
blk.4.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.4.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.4.ffn_gate_inp.weightF32[1024, 32]
-
blk.4.ffn_norm.weightF32[1024]
-
blk.4.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.5.attn_k.weightQ8_0[1024, 512]
-
blk.5.attn_norm.weightF32[1024]
-
blk.5.attn_output.weightQ8_0[1024, 1024]
-
blk.5.attn_q.weightQ8_0[1024, 1024]
-
blk.5.attn_v.weightQ8_0[1024, 512]
-
blk.5.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.5.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.5.ffn_gate_inp.weightF32[1024, 32]
-
blk.5.ffn_norm.weightF32[1024]
-
blk.5.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.6.attn_k.weightQ8_0[1024, 512]
-
blk.6.attn_norm.weightF32[1024]
-
blk.6.attn_output.weightQ8_0[1024, 1024]
-
blk.6.attn_q.weightQ8_0[1024, 1024]
-
blk.6.attn_v.weightQ8_0[1024, 512]
-
blk.6.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.6.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.6.ffn_gate_inp.weightF32[1024, 32]
-
blk.6.ffn_norm.weightF32[1024]
-
blk.6.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.7.attn_k.weightQ8_0[1024, 512]
-
blk.7.attn_norm.weightF32[1024]
-
blk.7.attn_output.weightQ8_0[1024, 1024]
-
blk.7.attn_q.weightQ8_0[1024, 1024]
-
blk.7.attn_v.weightQ8_0[1024, 512]
-
blk.7.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.7.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.7.ffn_gate_inp.weightF32[1024, 32]
-
blk.7.ffn_norm.weightF32[1024]
-
blk.7.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.8.attn_k.weightQ8_0[1024, 512]
-
blk.8.attn_norm.weightF32[1024]
-
blk.8.attn_output.weightQ8_0[1024, 1024]
-
blk.8.attn_q.weightQ8_0[1024, 1024]
-
blk.8.attn_v.weightQ8_0[1024, 512]
-
blk.8.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.8.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.8.ffn_gate_inp.weightF32[1024, 32]
-
blk.8.ffn_norm.weightF32[1024]
-
blk.8.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.9.attn_k.weightQ8_0[1024, 512]
-
blk.9.attn_norm.weightF32[1024]
-
blk.9.attn_output.weightQ8_0[1024, 1024]
-
blk.9.attn_q.weightQ8_0[1024, 1024]
-
blk.9.attn_v.weightQ8_0[1024, 512]
-
blk.9.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.9.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.9.ffn_gate_inp.weightF32[1024, 32]
-
blk.9.ffn_norm.weightF32[1024]
-
blk.9.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.10.attn_k.weightQ8_0[1024, 512]
-
blk.10.attn_norm.weightF32[1024]
-
blk.10.attn_output.weightQ8_0[1024, 1024]
-
blk.10.attn_q.weightQ8_0[1024, 1024]
-
blk.10.attn_v.weightQ8_0[1024, 512]
-
blk.10.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.10.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.10.ffn_gate_inp.weightF32[1024, 32]
-
blk.10.ffn_norm.weightF32[1024]
-
blk.10.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.11.attn_k.weightQ8_0[1024, 512]
-
blk.11.attn_norm.weightF32[1024]
-
blk.11.attn_output.weightQ8_0[1024, 1024]
-
blk.11.attn_q.weightQ8_0[1024, 1024]
-
blk.11.attn_v.weightQ8_0[1024, 512]
-
blk.11.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.11.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.11.ffn_gate_inp.weightF32[1024, 32]
-
blk.11.ffn_norm.weightF32[1024]
-
blk.11.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.12.attn_k.weightQ8_0[1024, 512]
-
blk.12.attn_norm.weightF32[1024]
-
blk.12.attn_output.weightQ8_0[1024, 1024]
-
blk.12.attn_q.weightQ8_0[1024, 1024]
-
blk.12.attn_v.weightQ8_0[1024, 512]
-
blk.12.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.12.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.12.ffn_gate_inp.weightF32[1024, 32]
-
blk.12.ffn_norm.weightF32[1024]
-
blk.12.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.13.attn_k.weightQ8_0[1024, 512]
-
blk.13.attn_norm.weightF32[1024]
-
blk.13.attn_output.weightQ8_0[1024, 1024]
-
blk.13.attn_q.weightQ8_0[1024, 1024]
-
blk.13.attn_v.weightQ8_0[1024, 512]
-
blk.13.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.13.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.13.ffn_gate_inp.weightF32[1024, 32]
-
blk.13.ffn_norm.weightF32[1024]
-
blk.13.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.14.attn_k.weightQ8_0[1024, 512]
-
blk.14.attn_norm.weightF32[1024]
-
blk.14.attn_output.weightQ8_0[1024, 1024]
-
blk.14.attn_q.weightQ8_0[1024, 1024]
-
blk.14.attn_v.weightQ8_0[1024, 512]
-
blk.14.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.14.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.14.ffn_gate_inp.weightF32[1024, 32]
-
blk.14.ffn_norm.weightF32[1024]
-
blk.14.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.15.attn_k.weightQ8_0[1024, 512]
-
blk.15.attn_norm.weightF32[1024]
-
blk.15.attn_output.weightQ8_0[1024, 1024]
-
blk.15.attn_q.weightQ8_0[1024, 1024]
-
blk.15.attn_v.weightQ8_0[1024, 512]
-
blk.15.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.15.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.15.ffn_gate_inp.weightF32[1024, 32]
-
blk.15.ffn_norm.weightF32[1024]
-
blk.15.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.16.attn_k.weightQ8_0[1024, 512]
-
blk.16.attn_norm.weightF32[1024]
-
blk.16.attn_output.weightQ8_0[1024, 1024]
-
blk.16.attn_q.weightQ8_0[1024, 1024]
-
blk.16.attn_v.weightQ8_0[1024, 512]
-
blk.16.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.16.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.16.ffn_gate_inp.weightF32[1024, 32]
-
blk.16.ffn_norm.weightF32[1024]
-
blk.16.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.17.attn_k.weightQ8_0[1024, 512]
-
blk.17.attn_norm.weightF32[1024]
-
blk.17.attn_output.weightQ8_0[1024, 1024]
-
blk.17.attn_q.weightQ8_0[1024, 1024]
-
blk.17.attn_v.weightQ8_0[1024, 512]
-
blk.17.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.17.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.17.ffn_gate_inp.weightF32[1024, 32]
-
blk.17.ffn_norm.weightF32[1024]
-
blk.17.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.18.attn_k.weightQ8_0[1024, 512]
-
blk.18.attn_norm.weightF32[1024]
-
blk.18.attn_output.weightQ8_0[1024, 1024]
-
blk.18.attn_q.weightQ8_0[1024, 1024]
-
blk.18.attn_v.weightQ8_0[1024, 512]
-
blk.18.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.18.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.18.ffn_gate_inp.weightF32[1024, 32]
-
blk.18.ffn_norm.weightF32[1024]
-
blk.18.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.19.attn_k.weightQ8_0[1024, 512]
-
blk.19.attn_norm.weightF32[1024]
-
blk.19.attn_output.weightQ8_0[1024, 1024]
-
blk.19.attn_q.weightQ8_0[1024, 1024]
-
blk.19.attn_v.weightQ8_0[1024, 512]
-
blk.19.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.19.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.19.ffn_gate_inp.weightF32[1024, 32]
-
blk.19.ffn_norm.weightF32[1024]
-
blk.19.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.20.attn_k.weightQ8_0[1024, 512]
-
blk.20.attn_norm.weightF32[1024]
-
blk.20.attn_output.weightQ8_0[1024, 1024]
-
blk.20.attn_q.weightQ8_0[1024, 1024]
-
blk.20.attn_v.weightQ8_0[1024, 512]
-
blk.20.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.20.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.20.ffn_gate_inp.weightF32[1024, 32]
-
blk.20.ffn_norm.weightF32[1024]
-
blk.20.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.21.attn_k.weightQ8_0[1024, 512]
-
blk.21.attn_norm.weightF32[1024]
-
blk.21.attn_output.weightQ8_0[1024, 1024]
-
blk.21.attn_q.weightQ8_0[1024, 1024]
-
blk.21.attn_v.weightQ8_0[1024, 512]
-
blk.21.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.21.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.21.ffn_gate_inp.weightF32[1024, 32]
-
blk.21.ffn_norm.weightF32[1024]
-
blk.21.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.22.attn_k.weightQ8_0[1024, 512]
-
blk.22.attn_norm.weightF32[1024]
-
blk.22.attn_output.weightQ8_0[1024, 1024]
-
blk.22.attn_q.weightQ8_0[1024, 1024]
-
blk.22.attn_v.weightQ8_0[1024, 512]
-
blk.22.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.22.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.22.ffn_gate_inp.weightF32[1024, 32]
-
blk.22.ffn_norm.weightF32[1024]
-
blk.22.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
blk.23.attn_k.weightQ8_0[1024, 512]
-
blk.23.attn_norm.weightF32[1024]
-
blk.23.attn_output.weightQ8_0[1024, 1024]
-
blk.23.attn_q.weightQ8_0[1024, 1024]
-
blk.23.attn_v.weightQ8_0[1024, 512]
-
blk.23.ffn_down_exps.weightQ8_0[512, 1024, 32]
-
blk.23.ffn_gate_exps.weightQ8_0[1024, 512, 32]
-
blk.23.ffn_gate_inp.weightF32[1024, 32]
-
blk.23.ffn_norm.weightF32[1024]
-
blk.23.ffn_up_exps.weightQ8_0[1024, 512, 32]
-
output_norm.weightF32[1024]
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