Qwen3.8-Flash-Next FP4-Local QAD on TensorFold v1.1.4, one Spark, 262K
1× DGX Spark · TensorFold ·
Tested configuration
- Hardware
- 1x NVIDIA DGX Spark (128GB)
- Serving runtime
- TensorFold
- Runtime version or build
- TensorFold v1.1.4 (eab85828) with the recipe's patches 0001-0017
- Weight quantization or precision
- FP4-Local QAD
- Benchmark date
- Recipe credit
- MiaAI-Lab Qwen3.8 Flash Next single-Spark TensorFold recipeBy MiaAI-Lab
20e15e41f2f514a0d287710f4665ee1b5009ba41
Decode sweep
~256 input / 512 output tokens per request. Output throughput is aggregate across concurrent requests. TTFT means time to first token.
| Concurrency | Output tok/s | TTFT p95 (s) |
|---|---|---|
| 1 | 41.54 | 0.35 |
| 2 | 83.18 | 0.42 |
| 4 | 111.73 | 0.97 |
| 8 | 98.64 | 19.84 |
| 12 | 95.15 | 44.68 |
| 16 | 88.23 | 69.13 |
Prefill and context
- Configured request window
- 262,144 tokens
- Longest tested prompt
- 245,760 target input tokens
- Validated retrieval
- ≈240K input tokens · passkey passed
Input throughput uses full request wall time. The passkey measures retrieval from one long prompt.
| Workload (c1) | Input tok/s | TTFT p95 (s) |
|---|---|---|
| 4,096 input tokens | 1,281.74 | 1.95 |
| 32,768 input tokens | 2,059.50 | 14.30 |
| 131,072 input tokens | 2,066.39 | 62.05 |
| 245,760 input tokensPasskey prompt | 1,936.59 | 126.64 |
Sustained load
20-minute soak at concurrency 8 · ~1024 input / 512 output tokens per request.
- Aggregate throughput
- 77.79 output tok/s
- TTFT p95
- 32.12 s
- Completed requests
- 189
- Request errors
- 0
Serving regression checks 13/13 passed
Tool-use evaluation
Complete · 100% completionSafety gate failed
Compare only against complete results with 100% completion under the same protocol ID.
- Protocol ID
tool-eval-standard-v1-b60e2f0- Workload
- 69 mock-tool scenarios per trial · 3 trials
- Mean score
- 89.0 / 100
- Pass@3 · passed at least once
- 84.1%
- Pass³ · passed every trial
- 84.1%