Qwen2.5-Coder-14B-Instruct VRAM & Memory Requirements
Specific checkpoints with sourced architecture and weight files. GGUF Q4_K_M (Qwen): 10.00 GiB–10.64 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-14B-Instruct - Revision
aedcc2d42b622764e023cf882b6652e646b95671 - Total parameters14,770,033,664
- Layers48
- 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
- 8.37 GiB (8,988,110,400 bytes)
- Raw KV cache
- 1.50 GiB (1,610,612,736 bytes)
- Engine extras / workspace
- 0.13 GiB–0.77 GiB
- Total estimated demand
- 10.00 GiB–10.64 GiB
Published artifacts
| Quantization | Selected files | Bytes | Engine | Evidence |
|---|---|---|---|---|
| BF16 safetensors bf16 | 29,540,067,328 (27.51 GiB / 29.54 GB) | vllm, sglang, unspecified | file-metadata | |
| GGUF Q4_K_M (Qwen) gguf-q4_k_m | qwen2.5-coder-14b-instruct-q4_k_m-00001-of-00002.gguf, qwen2.5-coder-14b-instruct-q4_k_m-00002-of-00002.gguf | 8,988,110,400 (8.37 GiB / 8.99 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q5_K_M (Qwen) gguf-q5_k_m | qwen2.5-coder-14b-instruct-q5_k_m-00001-of-00002.gguf, qwen2.5-coder-14b-instruct-q5_k_m-00002-of-00002.gguf | 10,508,873,280 (9.79 GiB / 10.51 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q6_K (Qwen) gguf-q6_k | qwen2.5-coder-14b-instruct-q6_k-00001-of-00002.gguf, qwen2.5-coder-14b-instruct-q6_k-00002-of-00002.gguf | 12,124,683,840 (11.29 GiB / 12.12 GB) | llama.cpp, ollama, unspecified | file-metadata |
| GGUF Q8_0 (Qwen) gguf-q8_0 | qwen2.5-coder-14b-instruct-q8_0-00001-of-00002.gguf, qwen2.5-coder-14b-instruct-q8_0-00002-of-00002.gguf | 15,701,597,760 (14.62 GiB / 15.70 GB) | llama.cpp, ollama, unspecified | file-metadata |
| INT4 theoretical | No published file | 7,385,016,832 (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 | 28.51 GiB–30.40 GiB | 29.64 GiB–31.56 GiB | 34.14 GiB–36.19 GiB |
| GGUF Q4_K_M (Qwen) | 8.87 GiB–9.42 GiB | 10.00 GiB–10.64 GiB | 14.50 GiB–15.46 GiB |
| GGUF Q5_K_M (Qwen) | 10.29 GiB–10.87 GiB | 11.41 GiB–12.09 GiB | 15.91 GiB–16.90 GiB |
| GGUF Q6_K (Qwen) | 11.79 GiB–12.40 GiB | 12.92 GiB–13.62 GiB | 17.42 GiB–18.44 GiB |
| GGUF Q8_0 (Qwen) | 15.12 GiB–15.80 GiB | 16.25 GiB–17.02 GiB | 20.75 GiB–21.84 GiB |
| INT4 theoretical Theoretical rows are not downloadable artifacts. | 7.38 GiB–9.87 GiB | 8.50 GiB–11.28 GiB | 13.00 GiB–16.82 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: 1,610,612,736 bytes (1.5000 GiB)
- 32,768 tokens: 6,442,450,944 bytes (6.00 GiB)
Related pages
Sources
- Qwen2.5-Coder-14B-Instruct model card
- config.json — revision aedcc2d42b622764e023cf882b6652e646b95671
- safetensors index — metadata.total_size=29540067328
- Artifact repo Qwen/Qwen2.5-Coder-14B-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-14b-instruct-q4_k_m-00001-of-00002.gguf, qwen2.5-coder-14b-instruct-q4_k_m-00002-of-00002.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.