Instructions to use 4fhct4sd/ocular-qwen3-4b-mixed01 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use 4fhct4sd/ocular-qwen3-4b-mixed01 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 4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf 4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf 4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf 4fhct4sd/ocular-qwen3-4b-mixed01: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 4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf 4fhct4sd/ocular-qwen3-4b-mixed01: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 4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf 4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M

Use Docker

docker model run hf.co/4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M

LM Studio
Jan
Ollama
How to use 4fhct4sd/ocular-qwen3-4b-mixed01 with Ollama:
```
ollama run hf.co/4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M
```
Unsloth Desktop

How to use 4fhct4sd/ocular-qwen3-4b-mixed01 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf 4fhct4sd/ocular-qwen3-4b-mixed01: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": "4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use 4fhct4sd/ocular-qwen3-4b-mixed01 with Docker Model Runner:
```
docker model run hf.co/4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M
```

Lemonade

How to use 4fhct4sd/ocular-qwen3-4b-mixed01 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull 4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M

Run and chat with the model

lemonade run user.ocular-qwen3-4b-mixed01-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use 4fhct4sd/ocular-qwen3-4b-mixed01 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf 4fhct4sd/ocular-qwen3-4b-mixed01: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 4fhct4sd/ocular-qwen3-4b-mixed01:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use 4fhct4sd/ocular-qwen3-4b-mixed01 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf 4fhct4sd/ocular-qwen3-4b-mixed01: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 "4fhct4sd/ocular-qwen3-4b-mixed01: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"

ocular-qwen3-4b-mixed01

DoRA finetunes of Qwen3-4B on a personal corpus, merged and quantized. Private. Two ranks trained identically except for adapter capacity — the pair is the point.

file	size	what
`A-q4_k_m.gguf`	2.50 GB	rank 128, Q4_K_M
`A-q8_0.gguf`	4.28 GB	rank 128, Q8_0
`C-q4_k_m.gguf`	2.50 GB	rank 32, Q4_K_M
`C-q8_0.gguf`	4.28 GB	rank 32, Q8_0

Training

base      Qwen/Qwen3-4B
method    DoRA over a 4-bit NF4 base (QLoRA-style), all 7 projections
seq       2048, 1 epoch, bf16, paged_adamw_8bit
data      22.08M tokens, 9 sources
hardware  RTX 5090, ~3.8h each

source	share
rosetta (1,182 tasks x 991 languages)	40.1%
soup (personal Claude sessions, tool calls kept)	35.2%
graphsoup (same corpus, graph-traversal rearrangement)	13.0%
chatgpt (personal, incl. reasoning traces)	5.6%
faraday / redbook / maxwell / machiavelli / socrates	6.2%

The rank comparison

Holdout is cut along whole tasks and whole languages, verified to have zero overlapping (task, language) pairs with training.

	rank 128	rank 32
trainable	265M (6.19%)	67M (1.64%)
train_loss	1.36	1.50
unseen-language ppl	15.23 -> 1.87	15.23 -> 2.29
unseen-task ppl	17.57 -> 4.73	17.57 -> 4.82
gap widening	+0.7844	+0.6024

unseen-language items are tasks trained on in other languages; unseen-task items were never seen at all. The widening gap is the transfer signal.

Rank 32 captures 77% of it with a quarter of the parameters. The extra capacity shows up almost entirely in unseen-language — re-rendering known tasks. On genuinely novel tasks the two are tied (4.73 vs 4.82), and rank 128 fit the training set harder for that non-gain.

Known defects

Training records were concatenated with no separator, so block boundaries taught transitions between unrelated documents as if they were continuations. Conversations got partial cover from <|im_end|>; the ~6% that are plain documents got none.
Packing slices mid-record every 2048 tokens.
weight_decay = 0.0.
The 40% Rosetta share is probably too high. Asked "what is 17 times 23", the 1.7B sibling writes a C# program instead of answering 391.

Provenance

Trained 2026-08-21 on rented RTX 5090s; merged, converted and quantized on the same boxes. Trail in the source repo under runs/2026-08-21/.

Downloads last month: 4

GGUF

Model size

4B params

Architecture

qwen3

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for 4fhct4sd/ocular-qwen3-4b-mixed01

Base model

Qwen/Qwen3-4B-Base

Finetuned

Qwen/Qwen3-4B

Quantized

(346)

this model