Qwen2.5-Coder-32B-Instruct VRAM & Memory Requirements
Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (Qwen): 20.61 GiB–21.50 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.
This result is an estimate, not a measured trace.Not verifiedFile-backed weights
Publisher and provenance
- PublisherQwen
- Hugging Face
Qwen/Qwen2.5-Coder-32B-Instruct - Revision
381fc969f78efac66bc87ff7ddeadb7e73c218a7 - Total parameters32,763,876,352
- Layers64
- Query heads40
- KV heads8 (gqa)
- Head dimension128
- Recorded context capability32,768
- Completenesscomplete · full
Memory layers in this estimate
Artifact bytes, raw KV, engine extras, and total demand are separate. This is not a runtime trace.
- Weight artifact
- 18.49 GiB (19,851,336,128 bytes)
- Raw KV cache
- 2.00 GiB (2,147,483,648 bytes)
- Engine extras / workspace
- 0.13 GiB–1.01 GiB
- Total estimated demand
- 20.61 GiB–21.50 GiB
Published artifacts
| Quantization | Selected files | Bytes | Engine | Evidence |
|---|---|---|---|---|
| BF16 safetensors bf16 | 65,527,752,704 (61.03 GiB / 65.53 GB) | vllm, sglang, unspecified | file-metadata | |
| GGUF Q4_K_M (Qwen) gguf-q4_k_m | qwen2.5-coder-32b-instruct-q4_k_m-00001-of-00003.gguf, qwen2.5-coder-32b-instruct-q4_k_m-00002-of-00003.gguf, qwen2.5-coder-32b-instruct-q4_k_m-00003-of-00003.gguf | 19,851,336,128 (18.49 GiB / 19.85 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q5_K_M (Qwen) gguf-q5_k_m | qwen2.5-coder-32b-instruct-q5_k_m-00001-of-00003.gguf, qwen2.5-coder-32b-instruct-q5_k_m-00002-of-00003.gguf, qwen2.5-coder-32b-instruct-q5_k_m-00003-of-00003.gguf | 23,262,157,216 (21.66 GiB / 23.26 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q6_K (Qwen) gguf-q6_k | qwen2.5-coder-32b-instruct-q6_k-00001-of-00004.gguf, qwen2.5-coder-32b-instruct-q6_k-00002-of-00004.gguf, qwen2.5-coder-32b-instruct-q6_k-00003-of-00004.gguf, qwen2.5-coder-32b-instruct-q6_k-00004-of-00004.gguf | 26,886,154,816 (25.04 GiB / 26.89 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q8_0 (Qwen) gguf-q8_0 | 34,820,885,152 (32.43 GiB / 34.82 GB) | llama.cpp, ollama, unspecified | file-metadata | |
| INT4 theoretical | No published file | 16,381,938,176 (theoretical) | — | theoretical |
Memory by quantization and context
Concurrent sequences = 1. KV defaults to FP16. The range includes documented runtime extras for the listed engine. These figures use the same core as the interactive calculator.
| Quantization | 2,048 tokens | 8,192 tokens | 32,768 tokens |
|---|---|---|---|
| BF16 safetensors | 62.15 GiB–64.04 GiB | 63.65 GiB–65.59 GiB | 69.65 GiB–71.77 GiB |
| GGUF Q4_K_M (Qwen) | 19.11 GiB–19.87 GiB | 20.61 GiB–21.50 GiB | 26.61 GiB–27.92 GiB |
| GGUF Q5_K_M (Qwen) | 22.29 GiB–23.11 GiB | 23.79 GiB–24.74 GiB | 29.79 GiB–31.16 GiB |
| GGUF Q6_K (Qwen) | 25.66 GiB–26.55 GiB | 27.16 GiB–28.18 GiB | 33.16 GiB–34.60 GiB |
| GGUF Q8_0 (Qwen) | 33.05 GiB–34.09 GiB | 34.55 GiB–35.72 GiB | 40.55 GiB–42.14 GiB |
| INT4 theoretical Theoretical rows are not downloadable artifacts. | 15.88 GiB–20.81 GiB | 17.38 GiB–22.69 GiB | 23.38 GiB–30.07 GiB |
Raw KV contribution
Standard GQA/MHA raw KV is 2 × layers × kvHeads × headDim × tokens × sequences × bytes per element. This page uses KV heads (8), not query heads (40). Using query heads would overstate this cache by 5×.
- 8,192 tokens, 1 sequence, FP16: 2,147,483,648 bytes (2.00 GiB)
- 32,768 tokens: 8,589,934,592 bytes (8.00 GiB)
Related pages
Sources
- Qwen2.5-Coder-32B-Instruct model card
- config.json — revision 381fc969f78efac66bc87ff7ddeadb7e73c218a7
- safetensors index — metadata.total_size=65527752704
- Artifact repo Qwen/Qwen2.5-Coder-32B-Instruct-GGUF
Assumptions that affect this result
- Requested context is within the recorded native window of 32768 tokens.
- Full-resident weight estimates use total parameters, including inactive MoE experts.
- Weight bytes come from the selected weight files only, not from every file in the repository.
- Download file size is not peak GPU memory. Runtime layout, allocator padding, and KV are extra.
- Counted files: qwen2.5-coder-32b-instruct-q4_k_m-00001-of-00003.gguf, qwen2.5-coder-32b-instruct-q4_k_m-00002-of-00003.gguf, qwen2.5-coder-32b-instruct-q4_k_m-00003-of-00003.gguf.
- Context budget is the number of cached tokens per sequence (prompt plus reserved generation).
- Concurrent sequences share one model replica; weights are not multiplied by sequence count.
- KV dtype is independent of weight quantization unless you change it.
Calculator
Loading catalog snapshot…
Search or pick a catalog checkpoint to see memory demand.
Search or pick a catalog checkpoint to see memory demand.