Underdog-27B-1.1
Underdog-27B by Conway is a 27-billion-parameter assistant for conversation, writing, coding, and summarization. This release combines post training with compact 4-bit weights for local inference. We applied post training to this model to run better in Underdog for human users, based on Qwen3.8.
Choose GGUF for generation with Splash or a compatible llama.cpp runtime, or the MLX package for native Apple Silicon workflows.
Local performance
On an Apple M5 Max, Splash with DFlash2 delivered up to 130.1 generated tokens per second in our coding test.
| Task | MLX (MLX-LM) | MLX (Splash + DFlash2) | GGUF (llama.cpp) | GGUF (Splash + DFlash2) |
|---|---|---|---|---|
| Email Writing | 33.4 tokens/s | 80.8 tokens/s | 26.1 tokens/s | 78.1 tokens/s |
| Code Generation | 33.3 tokens/s | 130.1 tokens/s | 26.1 tokens/s | 112.7 tokens/s |
| Long-Text Summarization | 32.8 tokens/s | 69.4 tokens/s | 25.7 tokens/s | 61.5 tokens/s |
Measured on Apple M5 Max with 40 GPU cores and 48 GiB unified memory. All four configurations use the same checkpoint in their respective MLX and GGUF packages. Values are means of two runs per task with temperature 0, thinking disabled, a 512-token output limit, and no reused prompt prefix. Engines were measured separately. Input lengths were 43, 44, and 3,003 tokens; generated output lengths may differ. Model loading and warmup are excluded. Splash 1.1.0 uses DFlash2 and INT8 KV; MLX 0.32.0 / MLX-LM 0.31.3 and Prism llama.cpp b10685 use standard generation without a draft model. The llama.cpp measurements use Metal on macOS. The table reports generation throughput.
Run with Splash
On an Apple Silicon Mac, install Splash 1.1.0 or later, then run:
splash serve \
--model ConwayResearch/Underdog-27B-1.1:Q4_1 \
--language-only \
--default-reasoning-effort medium
The explicit :Q4_1 selector selects the GGUF; Splash prepares its compatible draft model automatically. Open the chat page shown by the service, or connect your application through its OpenAI-compatible API. For medium thinking in the chat page, select Medium in the thinking menu.
Run GGUF with llama.cpp on Windows (NVIDIA)
Download the latest Windows CUDA release of llama.cpp from its releases page (llama-*-bin-win-cuda-*-x64.zip and the matching cudart-llama-bin-win-cuda-*-x64.zip, extracted into one folder). Then download the DFlash2 draft model next to the GGUF from this repository:
hf download incoai/Qwen3.8-27B-DFlash2-GGUF Qwen3.8-27B-DFlash2-Q8_0.gguf --local-dir ./underdog-4bit
llama-server.exe -m .\underdog-4bit\Underdog-27B-Q4_1.gguf `
-md .\underdog-4bit\Qwen3.8-27B-DFlash2-Q8_0.gguf `
--spec-type draft-dflash --spec-draft-n-max 5 `
-ngl 99 -ngld 99 -fa on -c 16384 --jinja
Choose your model package
| Variant | Files to download | How to use |
|---|---|---|
| GGUF | Underdog-27B-Q4_1.gguf and LICENSE | Load with Splash or a compatible llama.cpp engine. For Windows with an NVIDIA GPU, use a compatible CUDA build of llama.cpp. |
| MLX | The complete mlx/ directory and LICENSE | Load the mlx/ directory with a compatible MLX runtime. |
Download only the variant used by your application. Keep the MLX configuration, tokenizer, chat template, and weights together.
hf download ConwayResearch/Underdog-27B-1.1 \
--include "mlx/*" --include "LICENSE" \
--local-dir ./underdog-4bit
For GGUF:
hf download ConwayResearch/Underdog-27B-1.1 \
--include "*.gguf" --include "LICENSE" \
--local-dir ./underdog-4bit