File size: 3,024 Bytes
b388051 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 | ---
license: other
license_name: qwen-community-1.0
license_link: LICENSE
base_model:
- Qwen/Qwen3.8-Flash-Next
pipeline_tag: text-generation
tags:
- slipstream
- dgx-spark
- nvfp4
- qwen
---
# Qwen3.8-Flash-Next-Slipstream
Qwen3.8-Flash-Next for the Slipstream inference engine: our own NVFP4 quantization of the official BF16 checkpoint (routed experts and n-gram tables in NVFP4 with calibrated W4A4 activation scales, the rest kept in higher precision, MXFP8 prefill copies), plus the vision tower. Runs on one Spark (TP1) or two (TP2).
These are **Slipstream weight files** (`weights.vbin` + `index.json` per directory), the engine's own pre-packed format. They run only with Slipstream; they are not safetensors and do not load in other runtimes. Built with Slipstream's converters at commit `20c2dd9` (2026-10-31).
Base model: [Qwen/Qwen3.8-Flash-Next](https://huggingface.co/Qwen/Qwen3.8-Flash-Next). Engine: [Slipstream](https://github.com/Interactify/slipstream) (engine code release coming soon). Hardware: two NVIDIA DGX Spark (GB10, 128 GB each) connected over their 200 GbE link.
## Contents
| Directory | What | Node |
|---|---|---|
| `qwen38fnc` | model weights for one Spark (TP1) | single Spark |
| `qwen38fnc_tp2_r0` | model weights, rank 0 (TP2) | first |
| `qwen38fnc_tp2_r1` | model weights, rank 1 (TP2) | second |
| `qwen38_vision` | vision tower (BF16) | first |
## Download
On the first Spark (rank 0, runs the HTTP server):
```bash
hf download Interactify/Qwen3.8-Flash-Next-Slipstream --local-dir ~/vmodels \
--include "qwen38fnc_tp2_r0/*" "qwen38_vision/*"
```
On the second Spark (rank 1):
```bash
hf download Interactify/Qwen3.8-Flash-Next-Slipstream --local-dir ~/vmodels --include "qwen38fnc_tp2_r1/*"
```
On a single Spark, download `qwen38fnc` (and `qwen38_vision`) instead:
```bash
hf download Interactify/Qwen3.8-Flash-Next-Slipstream --local-dir ~/vmodels --include "qwen38fnc/*" "qwen38_vision/*"
```
The tokenizer, chat template and generation config:
```bash
hf download Qwen/Qwen3.8-Flash-Next --exclude "*.safetensors" # on the first Spark
```
The directory names match the defaults in Slipstream's `server/models.json`, so no configuration is needed.
## Run
```bash
python server/slipstream.py serve qwen3.8-flash-next --profile balanced
```
The server speaks the OpenAI and Anthropic APIs on port 8080. See the Slipstream README for profiles, the router and engine settings.
## Measured on Slipstream
- Prefill: 2110 (0.9k) / 3160 (3.5k) / 3410 (13.8k) tokens/s.
- Attention projections in 8-bit (standard): 4-6 % better perplexity than all-4-bit, for about 8 % more weight bytes.
- tool-eval-bench: 88/100, median agent turn 2.5 s (92 scenarios).
## Notes
- Context up to 262,144 tokens.
- MTP drafts from the model's own MTP head.
## License
Qwen Community License 1.0, as the base model. Read it before commercial use: it has conditions for very large products and for Model-as-a-Service / AI work assistant businesses. See `LICENSE` and `NOTICE`.
|