Scores from different versions, task subsets, harnesses and scoring rules are not directly comparable. How results are reviewed
Documented results
24 selected tests| View ReportGeneric LLM | Qwen3.8-27B | WikiText128K configured context | 9.7297 perplexityWord perplexity · Perplexity · Lower is better | — | — | 1 × W7800Single GPU · 48 GB VRAM/GPUConfiguration
|
|---|---|---|---|---|---|---|
| View ReportGeneric LLM | Qwen3.8-27B | WikiText16 chunks of512 tokens1 concurrent request | 7.2 perplexityWord perplexity · Perplexity · Lower is better | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration
|
| View ReportBetterBench | Qwen3.8-27B-PARO-MXFP4-v2 | WikiText256K configured context | 5.97 perplexityWord perplexity · Served perplexity · Lower is better | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Qwen3.8-27B | WikiTextwikitext-2-raw-v1 full test; 145 nonoverlapping 2048-token chunks1 concurrent request | 7.0344 perplexityWord perplexity · Token-weighted perplexity · Lower is better | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | halogen-swift1.5-qwen3.8-flash-next-v2-abliterated | WikiTextRaw test split70,774tokens1 concurrent request | 3.229 perplexityWord perplexity · Teacher-forced perplexity · Lower is better | — | — | 1 × Ryzen AI Max+ 395Single GPU · Unified memory · 128 GB RAMConfiguration
|
| View ReportGeneric LLM | Qwen3.8-27B | WikiText128K configured context | 9.7296 perplexityWord perplexity · Perplexity · Lower is better | — | — | 1 × W7800Single GPU · 48 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Qwen3.8-27B | WikiText16 chunks of512 tokens1 concurrent request | 7.11 perplexityWord perplexity · Perplexity · Lower is better | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration
|
| View ReportGeneric LLM | Qwen3.8-27B | WikiTextwikitext-2-raw-v1 full test; 145 nonoverlapping 2048-token chunks1 concurrent request | 7.0449 perplexityWord perplexity · Token-weighted perplexity · Lower is better | — | — | 1 × R9700Single GPU · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 8.8e-3 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 0.0126 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 94.3 %Score · n = 100 · Top-1 token agreement | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 6.7e-3 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 9.7e-3 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 95.2 %Score · n = 100 · Top-1 token agreement | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 6.9e-3 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 9.9e-3 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 95.25 %Score · n = 100 · Top-1 token agreement | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 7e-3 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 0.01 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 95.21 %Score · n = 100 · Top-1 token agreement | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 0.011 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 0.0155 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 93.8 %Score · n = 100 · Top-1 token agreement | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
| View ReportCustom evaluation | Qwen3.8-27B | WikiText96 x 500-character chunks | 0.0296 natsScore · KL divergence on reference top-K · Lower is better | — | — | 2 × R9700Tensor parallel · 32 GB VRAM/GPUConfiguration
|
Versions and harnesses
Versions: Not reported
Harnesses: BetterBench 0.6.0, Custom prompt_logprobs perplexity collector, Halogen engine ppl mode, kld.py
This is a selection across sources, not a ranking or a matched experiment. Reading benchmark reports
GPU benchmark pages: W7800 · R9700 · Ryzen AI Max+ 395