Instructions to use kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-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 kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-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 kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP
# Run inference directly in the terminal:
llama cli -hf kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP
# Run inference directly in the terminal:
llama cli -hf kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP
# Run inference directly in the terminal:
./build/bin/llama cli -hf kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

Use Docker

docker model run hf.co/kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

Llama new
LM Studio
Jan

vLLM

How to use kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-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": "kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

Ollama
How to use kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF with Ollama:
```
ollama run hf.co/kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP
```
Unsloth Desktop

How to use kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

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": "kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF with Docker Model Runner:
```
docker model run hf.co/kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP
```

Lemonade

How to use kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

Run and chat with the model

lemonade run user.KnowLine-31B-Gen1.5-ROCmFP4-GGUF-Q4_0_ROCMFP

List all available models

lemonade list

Hermes Agent

How to use kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-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 kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

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 kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP

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 "kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF:Q4_0_ROCMFP" \
  --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"

KnowLine-31B-Gen1.5 — ROCmFP4 GGUF

ROCmFP4 quantization of PelaAI/KnowLine-31B-Gen1.5 (Gemma 4 31B Instruct family, 31.27B params, Apache-2.0). Same recipe as the Gemma 4 31B Instruct ROCmFP4 build: Q4_0_ROCMFP4_COHERENT (ftype 102) with token_embd as q6_K, plus the BF16 vision projector and the official Gemma 4 31B Instruct MTP drafter.

Built for AMD Strix Halo (gfx1151) — Ryzen AI MAX+ 395, 128 GB unified memory. These files need a llama.cpp build that implements Q4_0_ROCMFP4_*. They will not load in stock llama.cpp / Ollama / LM Studio.

Files

file	size (bytes)	sha256
`KnowLine-31B-Gen1.5-Q4_0_ROCMFP4_COHERENT-imatrix-headQ6K.gguf`	17651000192	`245ae2c7a3fedf4a85129a500711e7195bac5de5dfb837435b777ac380d44700`
`mmproj-BF16.gguf`	1200726464	`bee558edda213750d7493cfa266734e6092b80885cc39d5aacbc21101f7c29ec`
`mtp-gemma-4-31B-it-Q8_0.gguf`	514687104	`5ae8b0117bed601e8924c6305bd5b0585de361d51f0e77091bcb4252cf1f27de`

Quant size 16818.21 MiB (4.60 BPW). token_embd.weight is q6_K. tie_word_embeddings = true — there is no output.weight; --output-tensor-type is a silent no-op and --token-embedding-type q6_K is the head flag.

mmproj-BF16.gguf is required for image input. The drafter is the base Gemma 4 31B Instruct assistant (Q8_0, 491 MiB).

How to run

Recommended text command (family seat flags; this is what produced the chat / tool-call check and the short MTP decode numbers):

llama-server -m KnowLine-31B-Gen1.5-Q4_0_ROCMFP4_COHERENT-imatrix-headQ6K.gguf \
  --spec-type draft-mtp --model-draft mtp-gemma-4-31B-it-Q8_0.gguf \
  --spec-draft-ngl 999 --spec-draft-n-max 5 \
  -ngl 999 -c 4096 -fa on -fit off --jinja --parallel 1

For vision, drop the --spec-type flags, pass --mmproj mmproj-BF16.gguf, and use -fa off. MTP and image input cannot be used together on this stack.

Measured

Hardware: AMD Ryzen AI MAX+ 395 (Strix Halo, gfx1151), 128 GB unified. Load: -ngl 999 -fa on -fit off.

Perplexity

llama-perplexity, corpus = kit holdout.txt (74993 bytes, sha256 167d6b72bcaf3e0139fa407889b7b1aeee18e304ec5b5ce6135efb669b9647f8), n_ctx=4096, --chunks 5, -ngl 999 -fa on -fit off.

file	PPL	vs Q8_0
Q8_0 reference	89.9893	1.000
COHERENT (this file)	93.2653	1.0364

1.02× Q8_0 bar = 91.789. This file is 1.0364× Q8_0.

Decode with the drafter (short prompts, `-c 4096`)

Same seat flags as above. Prompt-eval / decode from server timings; draft acceptance from draft_n_accepted/draft_n.

prompt tokens	decode tok/s	drafts accepted	mean accepted length
2111	24.61	56/78 (0.718)	3.5
26	32.04	11/13 (0.846)	3.75

Mean accepted length on the load pass: 3.68. Long prompts (≈1k–16k) with MTP abort this llama.cpp server (MTP boundary missing / speculative batch fail). Use the no-drafter command for long context.

Prefill / decode without the drafter, means of 2, speed256 harness

-ngl 999 -c 32768 -fa on -fit off --parallel 1. max_tokens 256, temperature 0, stream on. Quiet box, load_max 2.21 / 2.08.

prompt	prefill tok/s	decode tok/s
p1k (1600 tok)	264.5	10.8
p8k (14064 tok)	230.4	10.2
p16k (28225 tok)	193.1	9.5

Code-prompt mean decode 11.65 tok/s. Family no-drafter decode on this class of hardware is 11.81 tok/s at -c 4096.

Chat and tools

With the recommended text command: one-word capital of Australia → Canberra (finish=stop). Tool request for weather in Paris → get_weather {"city":"Paris"} (finish=tool_calls). Tool-call suite: required 7/8 and auto 7/8 (hello is not a tool call on both runs). Six text prompts 6/6 identical across the two speed legs (empty visible content: the replies sat in the reasoning channel).

Prompt-cache restore with MTP is skipped (spec-boundary-mismatch). A long cached prompt plus MTP aborted the server. That is a server limitation on this stack, not a file defect.

Quantize

Converted from the upstream safetensors with convert_hf_to_gguf.py --outtype bf16, then:

llama-quantize --imatrix imatrix.gguf \
  --output-tensor-type q6_K --token-embedding-type q6_K \
  KnowLine-31B-Gen1.5-BF16.gguf out.gguf Q4_0_ROCMFP4_COHERENT 16

Verify token_embd.weight is q6_K in the finished file.

License

Apache-2.0, same as PelaAI/KnowLine-31B-Gen1.5. The drafter is from the Gemma 4 31B Instruct assistant and remains under the Gemma terms of use.

Downloads last month: 185

GGUF

Model size

31B params

Architecture

gemma4

Hardware compatibility

4-bit

8-bit

Model tree for kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF

Base model

google/gemma-4-31B

Finetuned

google/gemma-4-31B-it

Finetuned

PelaAI/KnowLine-31B-Gen1.5

Quantized

(1)

this model

KnowLine-31B-Gen1.5 — ROCmFP4 GGUF

Files

How to run

Measured

Perplexity

Decode with the drafter (short prompts, -c 4096)

Prefill / decode without the drafter, means of 2, speed256 harness

Chat and tools

Quantize

License

Model tree for kingjones777/KnowLine-31B-Gen1.5-ROCmFP4-GGUF

Decode with the drafter (short prompts, `-c 4096`)