llm-architectures-explained

/models

Qwen3 0.6B

Alibaba Cloud (Qwen) · Qwen3 · open weights

Facts and where they come from

Released2025-04config.jsonconfig.jsonHugging Face repository creation date (api.createdAt)
Licenceapache-2.0config.jsonconfig.jsonREADME metadata: license
Total parameters600Mlabmodel cardREADME: Number of Parameters: 0.6B
Active parametersnot disclosednot disclosed
Context length32K tokenslabmodel cardREADME: Context Length: 32,768
Norm placementprecodemodelling codetransformers 5.18.0 / repo modelling code for qwen3: input_layernorm before attention, post_attention_layernorm before the MLP
Norm typeRMSNormcodemodelling codetransformers 5.18.0 / repo modelling code for qwen3: input_layernorm before attention, post_attention_layernorm before the MLP
QK-normyescodemodelling codetransformers 5.18.0 qwen3: q_norm and k_norm
Positional encodingRoPEcodemodelling coderotary on the full head (default)
Parallel attention and MLPnocodemodelling codetransformers 5.18.0 / repo modelling code for qwen3: input_layernorm before attention, post_attention_layernorm before the MLP

Architecture, drawn from the data

GQA 16q/8kv. Each column is one layer: its token mixer above, its feed-forward block below. Paler columns reuse another layer’s keys and values.

Qwen3 0.6B: layer stack and blockslayers (28)mixer / FFNlayer 0: GQA: 16 query / 8 KV heads · head 128layer 0: Gated MLP: 3072layer 1: GQA: 16 query / 8 KV heads · head 128layer 1: Gated MLP: 3072layer 2: GQA: 16 query / 8 KV heads · head 128layer 2: Gated MLP: 3072layer 3: GQA: 16 query / 8 KV heads · head 128layer 3: Gated MLP: 3072layer 4: GQA: 16 query / 8 KV heads · head 128layer 4: Gated MLP: 3072layer 5: GQA: 16 query / 8 KV heads · head 128layer 5: Gated MLP: 3072layer 6: GQA: 16 query / 8 KV heads · head 128layer 6: Gated MLP: 3072layer 7: GQA: 16 query / 8 KV heads · head 128layer 7: Gated MLP: 3072layer 8: GQA: 16 query / 8 KV heads · head 128layer 8: Gated MLP: 3072layer 9: GQA: 16 query / 8 KV heads · head 128layer 9: Gated MLP: 3072layer 10: GQA: 16 query / 8 KV heads · head 128layer 10: Gated MLP: 3072layer 11: GQA: 16 query / 8 KV heads · head 128layer 11: Gated MLP: 3072layer 12: GQA: 16 query / 8 KV heads · head 128layer 12: Gated MLP: 3072layer 13: GQA: 16 query / 8 KV heads · head 128layer 13: Gated MLP: 3072layer 14: GQA: 16 query / 8 KV heads · head 128layer 14: Gated MLP: 3072layer 15: GQA: 16 query / 8 KV heads · head 128layer 15: Gated MLP: 3072layer 16: GQA: 16 query / 8 KV heads · head 128layer 16: Gated MLP: 3072layer 17: GQA: 16 query / 8 KV heads · head 128layer 17: Gated MLP: 3072layer 18: GQA: 16 query / 8 KV heads · head 128layer 18: Gated MLP: 3072layer 19: GQA: 16 query / 8 KV heads · head 128layer 19: Gated MLP: 3072layer 20: GQA: 16 query / 8 KV heads · head 128layer 20: Gated MLP: 3072layer 21: GQA: 16 query / 8 KV heads · head 128layer 21: Gated MLP: 3072layer 22: GQA: 16 query / 8 KV heads · head 128layer 22: Gated MLP: 3072layer 23: GQA: 16 query / 8 KV heads · head 128layer 23: Gated MLP: 3072layer 24: GQA: 16 query / 8 KV heads · head 128layer 24: Gated MLP: 3072layer 25: GQA: 16 query / 8 KV heads · head 128layer 25: Gated MLP: 3072layer 26: GQA: 16 query / 8 KV heads · head 128layer 26: Gated MLP: 3072layer 27: GQA: 16 query / 8 KV heads · head 128layer 27: Gated MLP: 307201427× 28normGQA: 16 query / 8 KV heads · head 128+normGated MLP: 3072+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)596M
Active per token (modelled)596M
Without embeddings and output head440M total, 440M active
Published weights (Hugging Face count)596M
KV cache per token, BF16 (layers that grow with context)112 KiB
KV cache + state at 32K tokens, BF163.5 GiB
Decode FLOPs per token at 4K context2.13 GFLOP
Prefill FLOPs for a 4K prompt5.53 TFLOP

KV cache against context

Qwen3 0.6B: KV cache bytes against context length101001,00010,000980 KiB9.5 MiB95 MiB950 MiBcontext (tokens)KV cache + state (BF16)Qwen3 0.6B

Compare with other models →

Every architecture field

FieldValueSource
d_model1,024config.jsonconfig.jsonhidden_size
vocab151,936config.jsonconfig.jsonvocab_size
tied_embeddingstrueconfig.jsonconfig.jsontie_word_embeddings
mixers.full.typeattncodemodelling codeattention block
mixers.full.heads16config.jsonconfig.jsonnum_attention_heads
mixers.full.kv_heads8config.jsonconfig.jsonnum_key_value_heads
mixers.full.head_dim128config.jsonconfig.jsonhead_dim
mixers.full.qk_normtruecodemodelling codetransformers 5.18.0 qwen3: q_norm and k_norm
ffns.dense.typedensecodemodelling codeMLP block
ffns.dense.d_ff3,072config.jsonconfig.jsonintermediate_size
ffns.dense.gatedtruecodemodelling codeMLP: gated (SwiGLU/GeGLU)
layout28× full/denseconfig.jsonconfig.jsonnum_hidden_layers

Sources

Listed in the LLM Architecture Gallery checklist as “Qwen3 (0.6B)” (name only; see about).