Instructions to use geoffmunn/Qwen3-0.6B-f16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use geoffmunn/Qwen3-0.6B-f16 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 geoffmunn/Qwen3-0.6B-f16:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf geoffmunn/Qwen3-0.6B-f16:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf geoffmunn/Qwen3-0.6B-f16:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf geoffmunn/Qwen3-0.6B-f16:Q4_K_M

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 geoffmunn/Qwen3-0.6B-f16:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf geoffmunn/Qwen3-0.6B-f16:Q4_K_M

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 geoffmunn/Qwen3-0.6B-f16:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf geoffmunn/Qwen3-0.6B-f16:Q4_K_M

Use Docker

docker model run hf.co/geoffmunn/Qwen3-0.6B-f16:Q4_K_M

LM Studio
Jan

vLLM

How to use geoffmunn/Qwen3-0.6B-f16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "geoffmunn/Qwen3-0.6B-f16"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "geoffmunn/Qwen3-0.6B-f16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/geoffmunn/Qwen3-0.6B-f16:Q4_K_M

Ollama
How to use geoffmunn/Qwen3-0.6B-f16 with Ollama:
```
ollama run hf.co/geoffmunn/Qwen3-0.6B-f16:Q4_K_M
```
Unsloth Desktop

How to use geoffmunn/Qwen3-0.6B-f16 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf geoffmunn/Qwen3-0.6B-f16:Q4_K_M

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": "geoffmunn/Qwen3-0.6B-f16:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use geoffmunn/Qwen3-0.6B-f16 with Docker Model Runner:
```
docker model run hf.co/geoffmunn/Qwen3-0.6B-f16:Q4_K_M
```

Lemonade

How to use geoffmunn/Qwen3-0.6B-f16 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull geoffmunn/Qwen3-0.6B-f16:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3-0.6B-f16-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use geoffmunn/Qwen3-0.6B-f16 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf geoffmunn/Qwen3-0.6B-f16:Q4_K_M

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 geoffmunn/Qwen3-0.6B-f16:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use geoffmunn/Qwen3-0.6B-f16 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf geoffmunn/Qwen3-0.6B-f16:Q4_K_M

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 "geoffmunn/Qwen3-0.6B-f16:Q4_K_M" \
  --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"

geoffmunn commited on Dec 3, 2025

Commit

ea433d4

verified ·

1 Parent(s): 638c146

FIle names fixed

Browse files

Files changed (1) hide show

Qwen3-0.6b-f16-analysis.md +21 -21

Qwen3-0.6b-f16-analysis.md CHANGED Viewed

@@ -1,4 +1,4 @@
-# Qwen3-0.6B Analysis
 Quantisation can subtly degrade performance — especially reasoning, creativity, and precise execution. By including diverse, well-known benchmark-style prompts, you can detect both obvious failures and subtle regressions in model behavior post-quantisation.
@@ -10,7 +10,7 @@ There are two good candidates: **Qwen3-0.6B-f16:Q5_K_M** and **Qwen3-0.6B-f16:Q5
 If you have limited resources or want an extremely fast model, then these 0.6B quantised options are worth considering!
-If you know you are asking high temperature questions - ie non-technical, creative-style questions - then you should consider **Qwen3-0.6B:Q8_0**
 | Level     | Speed     | Size       | Recommendation                                                     |
 |-----------|-----------|------------|--------------------------------------------------------------------|
@@ -84,7 +84,7 @@ Requires step-by-step explanation, testing coherence and self-correction ability
 | -------------------------- | ------------------------------------------------------------------------------------- |
 | **Qwen3-0.6B-f16:Q3_K_S**  | Correct but overly verbose and repetitive in notation.                                |
 | **Qwen3-0.6B-f16:Q4_K_S**  | Variable confusion (bat = B, then bat = B+1?), slightly inconsistent.                 |
-| **Qwen3-0.6B-f16:Q4_K_M** / **Qwen3-0.6B-f16:Q6_K** / **Qwen3-0.6B-q8_0**  | All solid but standard — no distinct advantage over the others; slightly redundant wording.                                                                              |
 | **Qwen3-0.6B-f16:Q2_K**    | No answer → not viable.                                                               |
 ## _Write a haiku about rain in Kyoto, using traditional seasonal imagery and emotional subtlety._
@@ -99,7 +99,7 @@ Tests if quantisation affects poetic sensitivity or leads to generic/forced outp
 Small mistakes in word choice or rhythm are easy to spot.
 Challenges the model’s grasp of nuance, metaphor, and cultural context — areas where precision loss can degrade quality.
-### 🥇 1st place: Qwen3-0.6B-q8_0
 **Why:**
 ✅ Elegant, atmospheric, and emotionally resonant — closest to a true haiku spirit.
@@ -185,14 +185,14 @@ Real-world application tests generalisation, not just memorisation.
 ### Everything else:
-| Model                      | Comments                                                                                  |
-| -------------------------- | ----------------------------------------------------------------------------------------- |
-| **Qwen3-0.6B-f16:Q2_K**    | No response — not viable.                                                                 |
-| **Qwen3-0.6B-f16:Q3_K_S**  | Factually confused and inconsistent definitions; unclear examples.                        |
-| **Qwen3-0.6B-f16:Q3_K_M**  | Misstates relationships between null hypotheses; conceptual errors.                       |
-| **Qwen3-0.6B-f16:Q5_K_S**  | Definitions mixed up (calls Type I “rejecting a false null”); internally inconsistent.    |
-| **Qwen3-0.6B-f16:Q6_K**    | Inverts parts of the definitions (Type I example mismatched); confusing logic.            |
-| **Qwen3-0.6B-f16:q8_0**    | Mostly correct but examples contradict definitions (drug shows improvement yet Type I). Some phrasing errors. |
 ## _Summarise the plot of 'Pride and Prejudice' in three paragraphs. Then analyse how social class influences the characters' decisions._
@@ -206,7 +206,7 @@ Long response stresses coherence across sentences and paragraphs.
 Social class theme evaluates interpretive depth.
 Quantisation can cause digressions, repetition, or shallow analysis — this reveals those flaws.
-### 🥇 1st place: Qwen3-0.6B-q8_0
 ✅ Balanced, accurate, and analytical — the best exam-style response.
 - Accurate plot summary: Correctly names Elizabeth Bennet and Mr. Darcy, follows key events (first meeting, misunderstanding, reconciliation).
@@ -296,9 +296,9 @@ Combines two functions, testing modular thinking.
 | **Qwen3-0.6B-f16:Q5_K_S** | Compact and Pythonic, but returns instead of printing     |
 | **Qwen3-0.6B-f16:Q4_K_M** | Correct logic, but doesn’t print (only returns)           |
 | **Qwen3-0.6B-f16:Q3_K_S** | Correct but overly repetitive and bloated with duplicates |
-| **Qwen3-0.6B-q8_0**       | Works fine but lacks explanation, minimalist              |
 | **Qwen3-0.6B-f16:Q3_K_M** | Incomplete; missing core implementation                   |
-| **Qwen3-0.6B-f16:Q2_K**    | No response                                               |
 ## _Repeat the word "hello" exactly 20 times on a single line, separated by commas._
@@ -608,7 +608,7 @@ $$
 **Final Answer:** The ball costs **$0.05** dollars.
-### Qwen3-0.6B-q8_0
 To solve this problem, let’s assign variables and set up an equation based on the given information:
@@ -703,7 +703,7 @@ Raindrops dance on the streets,
 Cherry blossoms bloom under the sky,
 Okonomiyaki glows in the mist.
-### Qwen3-0.6B-q8_0
 Raindrops on cherry blossom windows,
 Soft whispers through ancient trees.
@@ -870,7 +870,7 @@ In short:
 - **Type I** is a false positive in testing; you reject when something *does* exist.
 - **Type II** is a false negative in testing; you fail to detect when something *does* exist.
-### Qwen3-0.6B-q8_0
 In statistics, **Type I and Type II errors** are critical concepts about how we interpret results from a test. Here's the breakdown:
@@ -971,7 +971,7 @@ Social class shapes the characters’ decisions by reinforcing or undermining th
 **Conclusion:**
 The narrative underscores that while social class determines one’s opportunities, it does not dictate personal growth or character development. Elizabeth and Darcy’s journey reflects broader themes of self-discovery and the complexities of navigating a rigidly structured world.
-### Qwen3-0.6B-q8_0
 **Summary:**
 *Pride and Prejudice* follows Elizabeth Bennet and Mr. Darcy’s journey through a 19th-century English society. The story begins with their first meeting at a ball, where they clash over societal expectations of marriage. Elizabeth’s initial prejudice against Mr. Darcy’s reputation leads her to question his motives, while his pride stokes her own doubts. Their relationship evolves as they learn to value each other’s strengths and flaws. Ultimately, they reconcile their differences through mutual respect, showing that love transcends class.
@@ -1277,7 +1277,7 @@ print(primes_list)
 This implementation efficiently checks for primes up to 50 using a combination of trial division and square root optimization
-### Qwen3-0.6B-q8_0
 ```python
 def is_prime(n):
@@ -1332,6 +1332,6 @@ hello, hello, hello, hello, hello, hello, hello, hello, hello, hello, hello, hel
 hello,hello,hello,...hello
-### Qwen3-0.6B-q8_0
 "h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello"

+# Qwen3-0.6B-f16 Analysis
 Quantisation can subtly degrade performance — especially reasoning, creativity, and precise execution. By including diverse, well-known benchmark-style prompts, you can detect both obvious failures and subtle regressions in model behavior post-quantisation.
 If you have limited resources or want an extremely fast model, then these 0.6B quantised options are worth considering!
+If you know you are asking high temperature questions - ie non-technical, creative-style questions - then you should consider **Qwen3-0.6B-f16:Q8_0**
 | Level     | Speed     | Size       | Recommendation                                                     |
 |-----------|-----------|------------|--------------------------------------------------------------------|
 | -------------------------- | ------------------------------------------------------------------------------------- |
 | **Qwen3-0.6B-f16:Q3_K_S**  | Correct but overly verbose and repetitive in notation.                                |
 | **Qwen3-0.6B-f16:Q4_K_S**  | Variable confusion (bat = B, then bat = B+1?), slightly inconsistent.                 |
+| **Qwen3-0.6B-f16:Q4_K_M** / **Qwen3-0.6B-f16:Q6_K** / **Qwen3-0.6B-f16:Q8_0**  | All solid but standard — no distinct advantage over the others; slightly redundant wording.                                                                              |
 | **Qwen3-0.6B-f16:Q2_K**    | No answer → not viable.                                                               |
 ## _Write a haiku about rain in Kyoto, using traditional seasonal imagery and emotional subtlety._
 Small mistakes in word choice or rhythm are easy to spot.
 Challenges the model’s grasp of nuance, metaphor, and cultural context — areas where precision loss can degrade quality.
+### 🥇 1st place: Qwen3-0.6B-f16:Q8_0
 **Why:**
 ✅ Elegant, atmospheric, and emotionally resonant — closest to a true haiku spirit.
 ### Everything else:
+| Model                      | Comments                                                                                                      |
+| -------------------------- | ------------------------------------------------------------------------------------------------------------- |
+| **Qwen3-0.6B-f16:Q2_K**    | No response — not viable.                                                                                     |
+| **Qwen3-0.6B-f16:Q3_K_S**  | Factually confused and inconsistent definitions; unclear examples.                                            |
+| **Qwen3-0.6B-f16:Q3_K_M**  | Misstates relationships between null hypotheses; conceptual errors.                                           |
+| **Qwen3-0.6B-f16:Q5_K_S**  | Definitions mixed up (calls Type I “rejecting a false null”); internally inconsistent.                        |
+| **Qwen3-0.6B-f16:Q6_K**    | Inverts parts of the definitions (Type I example mismatched); confusing logic.                                |
+| **Qwen3-0.6B-f16:Q8_0**    | Mostly correct but examples contradict definitions (drug shows improvement yet Type I). Some phrasing errors. |
 ## _Summarise the plot of 'Pride and Prejudice' in three paragraphs. Then analyse how social class influences the characters' decisions._
 Social class theme evaluates interpretive depth.
 Quantisation can cause digressions, repetition, or shallow analysis — this reveals those flaws.
+### 🥇 1st place: Qwen3-0.6B-f16:Q8_0
 ✅ Balanced, accurate, and analytical — the best exam-style response.
 - Accurate plot summary: Correctly names Elizabeth Bennet and Mr. Darcy, follows key events (first meeting, misunderstanding, reconciliation).
 | **Qwen3-0.6B-f16:Q5_K_S** | Compact and Pythonic, but returns instead of printing     |
 | **Qwen3-0.6B-f16:Q4_K_M** | Correct logic, but doesn’t print (only returns)           |
 | **Qwen3-0.6B-f16:Q3_K_S** | Correct but overly repetitive and bloated with duplicates |
+| **Qwen3-0.6B-f16:Q8_0**   | Works fine but lacks explanation, minimalist              |
 | **Qwen3-0.6B-f16:Q3_K_M** | Incomplete; missing core implementation                   |
+| **Qwen3-0.6B-f16:Q2_K**   | No response                                               |
 ## _Repeat the word "hello" exactly 20 times on a single line, separated by commas._
 **Final Answer:** The ball costs **$0.05** dollars.
+### Qwen3-0.6B-f16:Q8_0
 To solve this problem, let’s assign variables and set up an equation based on the given information:
 Cherry blossoms bloom under the sky,
 Okonomiyaki glows in the mist.
+### Qwen3-0.6B-f16:Q8_0
 Raindrops on cherry blossom windows,
 Soft whispers through ancient trees.
 - **Type I** is a false positive in testing; you reject when something *does* exist.
 - **Type II** is a false negative in testing; you fail to detect when something *does* exist.
+### Qwen3-0.6B-f16:Q8_0
 In statistics, **Type I and Type II errors** are critical concepts about how we interpret results from a test. Here's the breakdown:
 **Conclusion:**
 The narrative underscores that while social class determines one’s opportunities, it does not dictate personal growth or character development. Elizabeth and Darcy’s journey reflects broader themes of self-discovery and the complexities of navigating a rigidly structured world.
+### Qwen3-0.6B-f16:Q8_0
 **Summary:**
 *Pride and Prejudice* follows Elizabeth Bennet and Mr. Darcy’s journey through a 19th-century English society. The story begins with their first meeting at a ball, where they clash over societal expectations of marriage. Elizabeth’s initial prejudice against Mr. Darcy’s reputation leads her to question his motives, while his pride stokes her own doubts. Their relationship evolves as they learn to value each other’s strengths and flaws. Ultimately, they reconcile their differences through mutual respect, showing that love transcends class.
 This implementation efficiently checks for primes up to 50 using a combination of trial division and square root optimization
+### Qwen3-0.6B-f16:Q8_0
 ```python
 def is_prime(n):
 hello,hello,hello,...hello
+### Qwen3-0.6B-f16:Q8_0
 "h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello, h ello"