Instructions to use prism-ml/Ternary-Bonsai-2-27B-gguf with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use prism-ml/Ternary-Bonsai-2-27B-gguf with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16
# Run inference directly in the terminal:
llama cli -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16
# Run inference directly in the terminal:
llama cli -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16
# Run inference directly in the terminal:
./llama-cli -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Use Docker

docker model run hf.co/prism-ml/Ternary-Bonsai-2-27B-gguf:F16

LM Studio
Jan

vLLM

How to use prism-ml/Ternary-Bonsai-2-27B-gguf with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "prism-ml/Ternary-Bonsai-2-27B-gguf"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "prism-ml/Ternary-Bonsai-2-27B-gguf",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Ollama
How to use prism-ml/Ternary-Bonsai-2-27B-gguf with Ollama:
```
ollama run hf.co/prism-ml/Ternary-Bonsai-2-27B-gguf:F16
```
Unsloth Desktop

How to use prism-ml/Ternary-Bonsai-2-27B-gguf with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "prism-ml/Ternary-Bonsai-2-27B-gguf:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use prism-ml/Ternary-Bonsai-2-27B-gguf with Docker Model Runner:
```
docker model run hf.co/prism-ml/Ternary-Bonsai-2-27B-gguf:F16
```

Lemonade

How to use prism-ml/Ternary-Bonsai-2-27B-gguf with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Run and chat with the model

lemonade run user.Ternary-Bonsai-2-27B-gguf-F16

List all available models

lemonade list

Hermes Agent

How to use prism-ml/Ternary-Bonsai-2-27B-gguf with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use prism-ml/Ternary-Bonsai-2-27B-gguf with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf prism-ml/Ternary-Bonsai-2-27B-gguf:F16

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "prism-ml/Ternary-Bonsai-2-27B-gguf:F16" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

SYCL backend: any speculative type collapses performance (even target prefill drops ~200x) - draft model itself is healthy

#50

by Yoo00ooOO - opened 14 days ago

Discussion

Yoo00ooOO

14 days ago

Environment

GPU: Intel Arc A770 16GB (DG2/ACM-G10), driver 32.0.101.8991
Stack: oneAPI 2026.1, Level Zero backend (confirmed [level_zero:gpu:0] in banner)
OS: Windows 11, Python-free CLI build
Source: PrismML fork source snapshot (tarball, build banner reports b0-unknown)
Model: Ternary-Bonsai-2-27B PTQ1_0 (qwen35 hybrid arch, 1.75 bpw)
Draft: Qwen3.8-4B-Distill Q4_K_M (empero-ai, same vocab, qwen35 hybrid arch)

Summary

With any speculative type enabled (draft-simple, ngram-simple), performance
collapses far below the no-spec baseline — and notably the target model's own
prefill collapses too, which points at a synchronization/pacing problem in the
speculative driver rather than draft quality or GPU offload.

Measurements (same machine, same binaries)

Config	pp512	tg128 / generation
Target alone (`llama-bench`)	215 t/s	15.9 t/s
Draft alone (`llama-bench`, Qwen3.8-4B Q4_K_M)	1308 t/s	22.1 t/s
spec `draft-simple`, draft-n-max 16 (`llama-cli`)	5.9 t/s	0.5–0.9 t/s
spec `ngram-simple`, draft-n-max 16 (`llama-cli`, no draft model)	6.4 t/s	9.0 t/s

Notes:

The draft model is fully healthy in isolation: offloaded 34/34 layers to GPU,
1308 t/s prefill. Vocab is compatible (accepted tokens decode correctly).
Under draft-simple, verbose timings show draft_n: 6–13, draft_n_accepted: 1–3
(~20–30% accept rate) and predicted_per_token_ms of 1285–2183 ms — vs
~63 ms/token for the target alone. The per-step cost scales with draft_n, and
the draft alone should cost ~50 ms per drafted token at its standalone speed.
Under ngram-simple there is no draft model at all, yet target prefill
still drops from ~1300-class to 6.4 t/s. This is the strongest hint that the
spec driver enforces a per-step (or per-op) sync on the target queue on SYCL,
e.g. waiting on every verification batch instead of pipelining.

Reproduce

# baseline (healthy)
llama-bench -m Ternary-Bonsai-2-27B-PTQ1_0.gguf -ngl 999 -p 512 -n 128

# collapsed
llama-cli -m Ternary-Bonsai-2-27B-PTQ1_0.gguf \
  -md Qwen3.8-4B-Q4_K_M.gguf -ngl 999 -ngld 999 \
  --spec-type draft-simple --spec-draft-n-max 16 \
  -st -p "say OK" -n 32

# same collapse with no draft model
llama-cli -m Ternary-Bonsai-2-27B-PTQ1_0.gguf -ngl 999 \
  --spec-type ngram-simple --spec-draft-n-max 16 -st -p "say OK" -n 32

Secondary issue: default type inference crashes on plain drafts

Without an explicit --spec-type, a plain (non-sidecar) draft GGUF is routed
into the draft-mtp implementation and aborts:

speculative.cpp:2082: GGML_ASSERT(n_embd == llama_model_n_embd_out(...) &&
  "MTP input row width must match the target h_nextn width") failed

For a draft model with no MTP metadata, falling back to draft-simple (or
failing with a clear "no sidecar found, pass --spec-type" message) would be
friendlier than an assert.

Question

Is SYCL a supported backend for the rewritten speculative driver? If known-broken,
a docs note would save others the debugging session; if not known, happy to
provide verbose logs / tracing output on request.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment