Contributors › kyuz0's local-LLM benchmark configs
kyuz0's local-LLM benchmark configs
Every tracked benchmark config measured by kyuz0: model, quant, backend, tokens per second, source-linked and trust-tiered.
Snapshot 2026-10-29 · 413 configs on AI boxes and 288 on GPU cards · 121 models. Every number is generated from the tracker snapshot and linked to where it was measured; nothing here is typed by hand.
52 configs · 6 models · 2006 stars · 74.6 tok/s best single-stream · kyuz0 on github.
Configs measured by kyuz0
| Model | Hardware | Quant | Backend | Decode | Context | Trust | Source |
|---|---|---|---|---|---|---|---|
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/vulkan (speculative decode, mtp-3) | 74.6 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/vulkan (speculative decode, mtp-2) | 72.8 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm (speculative decode, mtp-3) | 68.3 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/vulkan-radv-mtp (speculative decode, mtp-3) | 65.6 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm (speculative decode, mtp-2) | 64.5 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/vulkan-radv-mtp (speculative decode, mtp-2) | 64.4 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/vulkan-radv | 62.8 tok/s (prefill 1242.6) | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/vulkan-radv-performance | 62.2 tok/s (prefill 1310.3) | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm-7.2.3-mtp (speculative decode, mtp-3) | 62.1 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm-7.2.3-mtp (speculative decode, mtp-2) | 58.9 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/vulkan (baseline) | 58.7 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/vulkan-radv-mtp (baseline) | 54.1 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm-7.14 | 53.5 tok/s (prefill 1604) | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm-7.14-pr26592 | 53 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm-7.2.4 | 51.7 tok/s (prefill 1383.8) | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q8_K_XL | llama.cpp/vulkan-radv | 49.2 tok/s (prefill 1176.6) | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q8_K_XL | llama.cpp/vulkan-radv-performance | 48.9 tok/s (prefill 1298.6) | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm (baseline) | 48.7 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q8_K_XL | llama.cpp/rocm-7.14 | 48.1 tok/s (prefill 1482.6) | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q8_K_XL | llama.cpp/rocm-7.14-pr26592 | 47.9 tok/s | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q8_K_XL | llama.cpp/rocm-7.2.4 | 46.5 tok/s (prefill 1215.1) | Structured table | source | |
| Qwen3.6-35B-A3B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm-7.2.3-mtp (baseline) | 46 tok/s | Structured table | source | |
| Qwen3.8-Flash-Next | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm (speculative decode, mtp) | 43.9 tok/s (prefill 1207) | Extracted, source-linked | source | |
| gemma-4-26B-A4B-it | Strix Halo | UD-Q8_K_XL | llama.cpp/rocm-7.2.4 | 41.8 tok/s (prefill 1699.7) | Structured table | source | |
| Qwen3.5-122B-A10B | Strix Halo | UD-Q4_K_XL | llama.cpp/rocm-7.2.4 | 21.6 tok/s (prefill 487.3) | Structured table | source |
More: all configs · hardware · news · contributors · methodology · llms.txt · llms-full.txt · API (OpenAPI) · JSON snapshot. Built by Altronis, private on-prem AI, Singapore.