Grouped by reported model identity. Exact weights, quantizations and software can differ. How results are reviewed
Documented results
9 selected tests| View ReportGeneric LLM | Qwen2.5-Coder-32B-Instruct | LoRA trainingDataset not reported | 0.4101 lossTraining loss · Loss · Lower is better | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration
|
|---|---|---|---|---|---|---|
| View Reportllama-bench | Qwen2.5-Coder-32B-Instruct | Token speed | — | 663.44 tok/sPrefill · mean · per request | — | 1 × W7900Single GPU · 48 GB VRAM/GPUConfiguration
|
| View ReportLocalMaxxing | Qwen2.5-Coder-32B-Instruct | Token speed | — | — | 16.24 tok/sGeneration · peak · unknown | 3 × R9700Multi-GPU · split unknown · 32 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Qwen2.5-Coder-32B-Instruct | TrainingSupervised fine-tuning | 112 tok/sTraining · whole system | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration
|
| View Reportllama-bench | Qwen2.5-Coder-32B-Instruct | Token speed | — | — | 23 tok/sGeneration · mean · per request | 1 × W7900Single GPU · 48 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Qwen2.5-Coder-32B-Instruct | TrainingSupervised fine-tuning | 137 tok/sTraining · range lower · whole system 142 tok/sTraining · range upper · whole system | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration
|
| View ReportServing benchmark | Qwen2.5-Coder-32B-Instruct | Token speedPython-Code-23k-ShareGPT / 50 requests | 26.75 tok/sTotal throughput · aggregate 21.29 tok/sThroughput · aggregate | — | — | 1 × W7900Single GPU · 48 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Qwen2.5-Coder-32B-Instruct | TrainingSupervised fine-tuning | 147 tok/sTraining · whole system | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration
|
| View ReportServing benchmark | Qwen2.5-Coder-32B-Instruct | Token speedPython-Code-23k-ShareGPT / 50 requests | 34.21 tok/sThroughput · aggregate 42.95 tok/sTotal throughput · aggregate | — | — | 1 × W7900Single GPU · 48 GB VRAM/GPUConfiguration
|
Quantizations and software
Quantizations: 4-bit QLoRA, Q4_K_M, Q8_0
Engines: LLaMA-Factory, llama.cpp, llama.cpp (hjc4869 fork)
This is a selection across sources, not a ranking or a matched experiment. Reading benchmark reports