GLM-5.3 Flash EXL3 4 bpw: TensorFold 1.1.9 with DFlash2, two Sparks, 1M
2× DGX Spark · TensorFold ·
Tested configuration
- Hardware
- 2x NVIDIA DGX Spark (128GB each)
- Serving runtime
- TensorFold
- Runtime version or build
- TensorFold 1.1.9 from f350cd3e plus the recipe's 110 patches
- Weight quantization or precision
- EXL3 4 bpw
- Benchmark date
- Recipe credit
- MiaAI-Lab GLM-5.3 Flash 1104 recipeBy MiaAI-Lab
13f2464ee84ab2b5dabf0bb264d1144b02cf5fb7
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 | 42.67 | 0.37 |
| 2 | 58.01 | 0.72 |
| 4 | 79.04 | 1.19 |
| 8 | 102.96 | 2.75 |
| 12 | 114.57 | 4.06 |
| 16 | 121.04 | 5.23 |
Prefill and context
- Configured request window
- 1,048,576 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 | 902.98 | 2.65 |
| 32,768 input tokens | 1,569.79 | 19.09 |
| 131,072 input tokens | 1,665.97 | 77.03 |
| 245,760 input tokensPasskey prompt | 622.77 | 391.26 |
Sustained load
20-minute soak at concurrency 8 · ~1024 input / 512 output tokens per request.
- Aggregate throughput
- 62.29 output tok/s
- TTFT p95
- 38.12 s
- Completed requests
- 149
- 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
- 93.0 / 100
- Pass@3 · passed at least once
- 91.3%
- Pass³ · passed every trial
- 91.3%