Qwen2.5-32B-Instruct Older
Where it runs 8K context · ≥ 5 t/s floor
| Setup | Best plan | Memory | Generate |
|---|---|---|---|
| 8 GB GPU + 32 GB RAM | Q4_K_Mllama.cpp CUDAGPU + CPU layers · 22/64 layers on GPU | KV unknown | ~1.5–1.8 t/s Estimated extrapolated |
| 12 GB GPU + 32 GB RAM | Q4_K_Mllama.cpp CUDAGPU + CPU layers · 36/64 layers on GPU | KV unknown | ~1.3–1.5 t/s Estimated extrapolated |
| 16 GB GPU + 64 GB RAM | AWQvLLMGPU streams weights from RAM · 4 GiB in RAM | KV unknown | ≤ 7.3 t/s Ceiling |
| 24 GB GPU + 64 GB RAM | Q5_K_Mllama.cpp CUDAFull GPU | KV unknown | ~25–31 t/s Estimated |
| 32 GB GPU + 128 GB RAM | Q5_K_Mllama.cpp CUDAFull GPU | KV unknown | ~39–62 t/s Estimated |
| 2× 24 GB GPUs + 128 GB RAM | Q8_0llama.cpp CUDALayer split across GPUs | KV unknown | ~15–20 t/s Estimated extrapolated |
| 96 GB workstation GPU + 256 GB RAM | FP16llama.cpp CUDAFull GPU | KV unknown | ~14–22 t/s Estimated |
| Mac, 24 GB | Does not fit | ||
| Mac, 48 GB | Q8_0llama.cpp MetalUnified memory | KV unknown | ~5.6–6.6 t/s Estimated |
| Mac, 128 GB | FP16llama.cpp MetalUnified memory | KV unknown | ~4.8–7.3 t/s Estimated |
| Mac, 512 GB | FP16llama.cpp MetalUnified memory | KV unknown | ~5.4–8.2 t/s Estimated |
| DGX Spark 128 GB | Q8_0llama.cpp CUDAUnified memory | KV unknown | ~6.0–6.6 t/s Estimated |
| Ryzen AI Max+ 395, 128 GB | Q5_K_Mllama.cpp VulkanUnified memory | KV unknown | ~6.0–8.7 t/s Estimated |
| CPU only, 64 GB DDR5-5600 dual channel | Q4_K_Mllama.cpp CPUCPU only | KV unknown | ~1.0–1.2 t/s Estimated extrapolated |
Same plan engine as the planner, with its default reserves and RAM. How plans are chosen.
Files exact Hugging Face bytes
| Artifact | Size | Files | Revision | Source |
|---|---|---|---|---|
| FP16 Qwen/Qwen2.5-32B-Instruct-GGUF | 61.04 GiB | 17 | a15e3cc10f | HF files |
| Q8_0 Qwen/Qwen2.5-32B-Instruct-GGUF | 32.43 GiB | 9 | a15e3cc10f | HF files |
| Q5_K_M Qwen/Qwen2.5-32B-Instruct-GGUF | 21.66 GiB | 6 | a15e3cc10f | HF files |
| Q4_K_M Qwen/Qwen2.5-32B-Instruct-GGUF | 18.49 GiB | 5 | a15e3cc10f | HF files |
| AWQ 4-bit Qwen/Qwen2.5-32B-Instruct-AWQ | 18.00 GiB | 5 | 5c7cb76a26 | HF files |
| GPTQ 4-bit Qwen/Qwen2.5-32B-Instruct-GPTQ-Int4 | 18.02 GiB | 5 | c83e67dfb2 | HF files |
| Native safetensors BF16 Qwen/Qwen2.5-32B-Instruct | 61.03 GiB | 17 | 5ede1c97bb | HF files |
Not found in the representative quant search (not proven unavailable): MLX 8-bit, MLX 4-bit, EXL3. Complete selected file sets; projectors, MTP/draft and lookup files are listed separately and budgeted only by plans that use them. File bytes are not allocated memory.
Quality
No Artificial Analysis evaluation mapped to this checkpoint.
Arena text rating
Unknown
Publisher-reported benchmarks
None extracted from the model card.
Open LLM Leaderboard (historical): 46.6 average · dataset. Historical Open LLM Leaderboard row (dataset last modified 2025-03-20); not 2026 coverage.
AA and Arena describe the hosted model in the stated mode, not each quantized file; scores are never inherited by base, pruned, abliterated or quantized variants and never averaged together.
How to run it
No model-specific recipe in the research. The planner binds the backend’s documented command template (llama.cpp, Ollama, LM Studio, MLX, vLLM) to your plan: open a plan’s details.
Recipes are source examples (settings are the publisher’s or author’s, not measured optima) and were not executed by LocalFit. A downloadable file does not prove your backend build supports the architecture.
Reported speeds 0
No reported measurement yet. Submit one.