---
license: apache-2.0
pipeline_tag: text-generation
tags: [agens, volundr, blockway, speculative-decoding, dflash2, draft-model]
new_version: Blockway/Agens-Volundr-32B-v1-DFlash2
---

> **Agens Volundr 32B v1 is out (2026-10-18):** the full release this Preview announced — agent post-training, 0/50 repetition loops, session continuation, INT4 / GGUF / drafter and a 16 GB Engine build. Use [Blockway/Agens-Volundr-32B-v1-DFlash2](https://huggingface.co/Blockway/Agens-Volundr-32B-v1-DFlash2) instead of this repository.

# Agens Volundr 32B Preview — DFlash2 drafter

**Main model card, benchmarks and discussion: [Blockway/Agens-Volundr-32B-Preview](https://huggingface.co/Blockway/Agens-Volundr-32B-Preview)**

A DFlash2 block drafter for speculative decoding with
[Agens Volundr 32B Preview](https://huggingface.co/Blockway/Agens-Volundr-32B-Preview). At each step it
drafts 8 tokens at once from the target model's hidden states (taps at layers 8, 21, 35, 53 and 67), and
the target verifies them in one pass. The target checks every token, so answers keep the target model's
quality; with greedy decoding the output can differ from non-speculative decoding only where two tokens
are numerically tied.

The drafter has 5 layers (about 1.9B parameters, 3.8 GB in bf16) and shares the target's embeddings and
output head, which are not stored here.

## Speed-up

Single user on two 48 GB GPUs, bf16, the Blockway sglang build, same server with and without the drafter:

| Output | Speed-up |
| --- | --- |
| JSON | 3.6× |
| Code | 2.0× |
| Replies with thinking on (code, Cantonese) | 1.6× |
| Cantonese chat | 1.3× |

On average 2.56 drafted tokens are accepted per step. The drafter is for single-user generation: the
drafter server runs 4 requests at a time, so with 8 or more concurrent users plain decoding gives more
total throughput.

## Use

Requires the Blockway sglang build (`ghcr.io/blockwayz/agens-sglang`, source at
`github.com/BlockWayz/agens-sglang`). Add these flags to the two-GPU bf16 command from the
[model card](https://huggingface.co/Blockway/Agens-Volundr-32B-Preview), and run the container with
`-e PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True -e VOLUNDR_FI_WORKSPACE_MB=256`. The larger FlashInfer workspace is needed on GPUs with many SMs, especially at tensor-parallel size 1 (the 2026-10-05 images default to 128 MB and can fail during CUDA-graph capture with `aligned_alloc ... but only 134217728 bytes available`):

```bash
    --speculative-algorithm DFLASH \
    --speculative-draft-model-path Blockway/Agens-Volundr-32B-Preview-DFlash2 \
    --speculative-num-draft-tokens 8 \
    --speculative-draft-model-quantization unquant \
    --mem-fraction-static 0.90 --context-length 16384 --chunked-prefill-size 2048 \
    --max-mamba-cache-size 4 --max-running-requests 4 --cuda-graph-max-bs 4
```

Speculative decoding needs `--attention-backend flashinfer` and serves text requests (leave out
`--enable-multimodal`).

## Licence

Apache-2.0. The licence files and `NOTICE` ship with the weights.