llm-architectures-explained

/models

OLMo 2 7B

Ai2 · OLMo · open weights

Facts and where they come from

Released2024-12config.jsonconfig.jsonHugging Face repository creation date (api.createdAt)
Licenceapache-2.0config.jsonconfig.jsonREADME metadata: license
Total parameters7Blabmodel cardmodel name OLMo-2-1124-7B
Active parametersnot disclosednot disclosed
Context length4K tokensconfig.jsonconfig.jsonmax_position_embeddings
Norm placementpostcodemodelling codetransformers 5.18.0 olmo2: post_attention_layernorm and post_feedforward_layernorm only
Norm typeRMSNormcodemodelling codetransformers 5.18.0 olmo2: post_attention_layernorm and post_feedforward_layernorm only
QK-normyescodemodelling codetransformers 5.18.0 olmo2: q_norm and k_norm
Positional encodingRoPEcodemodelling coderotary on the full head (default)
Parallel attention and MLPnocodemodelling codetransformers 5.18.0 olmo2: post_attention_layernorm and post_feedforward_layernorm only

Architecture, drawn from the data

MHA 32q/32kv. Each column is one layer: its token mixer above, its feed-forward block below. Paler columns reuse another layer’s keys and values.

OLMo 2 7B: layer stack and blockslayers (32)mixer / FFNlayer 0: MHA: 32 query / 32 KV heads · head 128layer 0: Gated MLP: 11008layer 1: MHA: 32 query / 32 KV heads · head 128layer 1: Gated MLP: 11008layer 2: MHA: 32 query / 32 KV heads · head 128layer 2: Gated MLP: 11008layer 3: MHA: 32 query / 32 KV heads · head 128layer 3: Gated MLP: 11008layer 4: MHA: 32 query / 32 KV heads · head 128layer 4: Gated MLP: 11008layer 5: MHA: 32 query / 32 KV heads · head 128layer 5: Gated MLP: 11008layer 6: MHA: 32 query / 32 KV heads · head 128layer 6: Gated MLP: 11008layer 7: MHA: 32 query / 32 KV heads · head 128layer 7: Gated MLP: 11008layer 8: MHA: 32 query / 32 KV heads · head 128layer 8: Gated MLP: 11008layer 9: MHA: 32 query / 32 KV heads · head 128layer 9: Gated MLP: 11008layer 10: MHA: 32 query / 32 KV heads · head 128layer 10: Gated MLP: 11008layer 11: MHA: 32 query / 32 KV heads · head 128layer 11: Gated MLP: 11008layer 12: MHA: 32 query / 32 KV heads · head 128layer 12: Gated MLP: 11008layer 13: MHA: 32 query / 32 KV heads · head 128layer 13: Gated MLP: 11008layer 14: MHA: 32 query / 32 KV heads · head 128layer 14: Gated MLP: 11008layer 15: MHA: 32 query / 32 KV heads · head 128layer 15: Gated MLP: 11008layer 16: MHA: 32 query / 32 KV heads · head 128layer 16: Gated MLP: 11008layer 17: MHA: 32 query / 32 KV heads · head 128layer 17: Gated MLP: 11008layer 18: MHA: 32 query / 32 KV heads · head 128layer 18: Gated MLP: 11008layer 19: MHA: 32 query / 32 KV heads · head 128layer 19: Gated MLP: 11008layer 20: MHA: 32 query / 32 KV heads · head 128layer 20: Gated MLP: 11008layer 21: MHA: 32 query / 32 KV heads · head 128layer 21: Gated MLP: 11008layer 22: MHA: 32 query / 32 KV heads · head 128layer 22: Gated MLP: 11008layer 23: MHA: 32 query / 32 KV heads · head 128layer 23: Gated MLP: 11008layer 24: MHA: 32 query / 32 KV heads · head 128layer 24: Gated MLP: 11008layer 25: MHA: 32 query / 32 KV heads · head 128layer 25: Gated MLP: 11008layer 26: MHA: 32 query / 32 KV heads · head 128layer 26: Gated MLP: 11008layer 27: MHA: 32 query / 32 KV heads · head 128layer 27: Gated MLP: 11008layer 28: MHA: 32 query / 32 KV heads · head 128layer 28: Gated MLP: 11008layer 29: MHA: 32 query / 32 KV heads · head 128layer 29: Gated MLP: 11008layer 30: MHA: 32 query / 32 KV heads · head 128layer 30: Gated MLP: 11008layer 31: MHA: 32 query / 32 KV heads · head 128layer 31: Gated MLP: 1100801631× 32MHA: 32 query / 32 KV heads · head 128norm+Gated MLP: 11008norm+full attentiondense FFN

Modelled costs

From the cost model, batch size 1. Totals the lab states are in the table above; differences come from rounding, from what a lab counts, or from parts the model does not describe (listed on the about page).

Parameters (modelled)7.3B
Active per token (modelled)7.3B
Without embeddings and output head6.48B total, 6.48B active
Published weights (Hugging Face count)7.3B
KV cache per token, BF16 (layers that grow with context)512 KiB
KV cache + state at 4K tokens, BF162 GiB
Decode FLOPs per token at 4K context15.9 GFLOP
Prefill FLOPs for a 4K prompt57.5 TFLOP

KV cache against context

OLMo 2 7B: KV cache bytes against context length101001,000980 KiB9.5 MiB95 MiB950 MiBcontext (tokens)KV cache + state (BF16)OLMo 2 7B

Compare with other models →

Every architecture field

FieldValueSource
d_model4,096config.jsonconfig.jsonhidden_size
vocab100,352config.jsonconfig.jsonvocab_size
tied_embeddingsfalseconfig.jsonconfig.jsontie_word_embeddings
mixers.full.typeattncodemodelling codeattention block
mixers.full.heads32config.jsonconfig.jsonnum_attention_heads
mixers.full.kv_heads32config.jsonconfig.jsonnum_key_value_heads
mixers.full.head_dim128codemodelling codetransformers 5.18.0: head_dim = hidden_size / num_attention_heads
mixers.full.qk_normtruecodemodelling codetransformers 5.18.0 olmo2: q_norm and k_norm
ffns.dense.typedensecodemodelling codeMLP block
ffns.dense.d_ff11,008config.jsonconfig.jsonintermediate_size
ffns.dense.gatedtruecodemodelling codeMLP: gated (SwiGLU/GeGLU)
layout32× full/denseconfig.jsonconfig.jsonnum_hidden_layers

Sources

Listed in the LLM Architecture Gallery checklist as “OLMo 2 (7B)” (name only; see about).