Contributors › hogeheer's local-LLM benchmark configs

hogeheer's local-LLM benchmark configs

Every tracked benchmark config measured by hogeheer: model, quant, backend, tokens per second, source-linked and trust-tiered.

Snapshot 2026-09-10 · 148 configs · 69 models. Every number is generated from the tracker snapshot and linked to where it was measured; nothing here is typed by hand.

37 configs · 25 models · 323 stars · 266 tok/s best single-stream · hogeheer on github.

Configs measured by hogeheer

ModelQuantBackendDecodeContextTrustSource
Qwen3 0.6BQ8_0llama.cpp/vulkan266 tok/s (prefill 13112)512Structured tablesource
Qwen3 0.6BQ8_0llama.cpp/rocm208.7 tok/s (prefill 4666.1)512Structured tablesource
LFM2.5 8B-A1BQ4_K_Mllama.cpp/vulkan176.5 tok/s (prefill 3398.4)512Structured tablesource
Qwen3-30B-A3B-Instruct-2507IQ4_XSllama.cpp/vulkan103.2 tok/s (prefill 1438.1)512Structured tablesource
Qwen3-Coder 30B-A3BQ4_K_Sllama.cpp/vulkan98 tok/s (prefill 1406.5)512Structured tablesource
Qwen3-Coder 30B-A3BUD-Q4_K_XLllama.cpp/vulkan97.1 tok/s (prefill 1400)512Structured tablesource
Qwen3-Coder 30B-A3BIQ4_XSllama.cpp/vulkan90.4 tok/s (prefill 1372.3)512Structured tablesource
Qwen3 30B-A3B NEO-MAXIQ4_XSllama.cpp/vulkan87.4 tok/s (prefill 1396.1)512Structured tablesource
Qwen3.6 35B-A3BQ4_0llama.cpp/vulkan81.3 tok/s (prefill 1243.5)512Structured tablesource
Nemotron Cascade 2 30B-A3BIQ4_XSllama.cpp/vulkan79 tok/s (prefill 1325.3)512Structured tablesource
Nemotron 3 Nano 30B-A3BIQ4_XSllama.cpp/vulkan76 tok/s (prefill 1312.5)512Structured tablesource
Qwen3.5 35B-A3BIQ4_XSllama.cpp/vulkan75.2 tok/s (prefill 1170.3)512Structured tablesource
Gemma 4 26B-A4B IT QATUD-Q4_K_XLllama.cpp/vulkan74.8 tok/s (prefill 1432)512Structured tablesource
Qwen3-Coder 30B-A3BUD-Q4_K_XLllama.cpp/rocm73.7 tok/s (prefill 1285.3)512Structured tablesource
Qwen AgentWorld 35B-A3BIQ4_XSllama.cpp/vulkan65.7 tok/s (prefill 1182.8)512Structured tablesource
Qwen3.5 35B-A3BQ4_K_Mllama.cpp/vulkan64.9 tok/s (prefill 1080)512Structured tablesource
Qwen3.6 35B-A3BQ4_K_Mllama.cpp/vulkan63.8 tok/s (prefill 1064)512Structured tablesource
Qwen3-Coder-Next 80B-A3BIQ4_XSllama.cpp/vulkan61.9 tok/s (prefill 739)512Structured tablesource
Qwen3.6 35B-A3BQ4_K_Mollama/vulkan60.6 tok/s (prefill 793.1)2048Structured tablesource
Qwen3-Next 80B-A3BUD-Q4_K_XLllama.cpp/vulkan54.9 tok/s (prefill 657)512Structured tablesource
Qwen3.5 35B-A3BQ4_K_Mllama.cpp/rocm54.7 tok/s (prefill 1047)512Structured tablesource
Gemma 4 26B-A4B ITQ4_K_Mllama.cpp/vulkan54.2 tok/s (prefill 1323.4)512Structured tablesource
Qwen3.6 35B-A3BQ4_K_Mllama.cpp/rocm52.7 tok/s (prefill 1186.2)512Structured tablesource
Qwen3-Next 80B-A3BUD-Q4_K_XLllama.cpp/rocm49.6 tok/s (prefill 800.4)512Structured tablesource
Gemma 4 26B-A4BQ4_K_Mllama.cpp/vulkan48.5 tok/s (prefill 1142)512Structured tablesource

More: all configs · hardware · methodology · llms.txt · llms-full.txt · API (OpenAPI) · JSON snapshot. Built by Altronis, private on-prem AI, Singapore.