Instructions to use jethrovic/bend-models with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use jethrovic/bend-models 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 jethrovic/bend-models:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf jethrovic/bend-models:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf jethrovic/bend-models:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf jethrovic/bend-models:Q4_K_M

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 jethrovic/bend-models:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf jethrovic/bend-models:Q4_K_M

Use Docker

docker model run hf.co/jethrovic/bend-models:Q4_K_M

Llama new
LM Studio
Jan
Ollama
How to use jethrovic/bend-models with Ollama:
```
ollama run hf.co/jethrovic/bend-models:Q4_K_M
```
Unsloth Desktop

How to use jethrovic/bend-models with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf jethrovic/bend-models: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": "jethrovic/bend-models:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use jethrovic/bend-models with Docker Model Runner:
```
docker model run hf.co/jethrovic/bend-models:Q4_K_M
```

Lemonade

How to use jethrovic/bend-models with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull jethrovic/bend-models:Q4_K_M

Run and chat with the model

lemonade run user.bend-models-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use jethrovic/bend-models with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf jethrovic/bend-models: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 jethrovic/bend-models:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use jethrovic/bend-models with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf jethrovic/bend-models: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 "jethrovic/bend-models: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"

bend-models / models /README.md

jethrovic

G30: models/ MANIFEST + SHA256SUMS + README (imatrix set)

b7c9d17 verified 16 days ago

preview code

Raw History Blame Contribute Delete

2.11 kB

# models/: imatrix-calibrated weights (G30, 2026-10-14)

Same quant types and byte budget as the shipped files, re-quantized with an importance matrix.
**Additive only:** the phone app still fetches `kev-4b-q4km/` at the repo root, and that dir is unchanged.

| file | bytes | sha256 | status |
|---|---|---|---|
| `kev-4b-q4km-imat/Kev-4B-Q4_K_M-imat.gguf` | 2,709,547,712 | `0535181cf3164a7271a7d3d1752a9ceab673c32bb316a081133ec238c186b51c` | ship (llama.cpp path); app NOT repointed |
| `kev-4b-iq4xs-imat/Kev-4B-IQ4_XS-imat.gguf` | 2,514,987,712 | `8a45e54a03a689368413de3bb6d33c1129ea416dfa7a8aade64c2f8c2c3b47dc` | published for llama.cpp; native path BLOCKED until gate receipt |
| `qwen38-27b-q4ks-imat-openjev-wide/Qwen3.8-27B-Q4_K_S-imat-openjev-wide.gguf` | 15,586,316,480 | `c1e505b2e4266d964cc75e8a801e1c54dfec1bd4205c6ea7072d15c9b907fc0f` | ship (board, llama-server); promotion into the board store is an owner action |
| `imatrix/kev4b-calib.imatrix.gguf` | 3,626,496 | `fe439522f2acc99fd040960f045f9582ccb89fe055d570ca883c61600a2f4737` | provenance |
| `imatrix/qwen38-27b-calib.imatrix.gguf` | 13,642,656 | `bd707218d6275dbcb342b23846776b306e535f431cb76f316349fa451138488c` | provenance |

## Fidelity (llama.cpp `--kl-divergence`, 40×512 ctx, 10,200 tokens, held-out lab prose)
| artifact | ref | mean KLD | top-1 % |
|---|---|---|---|
| Kev-4B Q4_K_M plain (shipped) | Kev BF16 | 0.04573 | 88.25 |
| **Kev-4B Q4_K_M-imat** | Kev BF16 | **0.02894** (−37%) | 89.81 |
| **Kev-4B IQ4_XS-imat** | Kev BF16 | **0.03593** | 89.22 |
| 27B Q4_K_S plain (shipped) | 27B Q6_K | 0.03652 | 90.84 |
| **27B Q4_K_S-imat** | 27B Q6_K | **0.02126** (−42%) | 93.30 |
The calibration data is in-domain (disjoint from the eval set), so expect a smaller gain on generic text.
The decision battery (40 scored questions) shows no regression vs shipped for any of these files.
**IQ4_XS-imat:** the native `llb` engine gate is still pending. Use it only with llama.cpp until that gate passes.
Verify: `sha256sum -c SHA256SUMS` inside `models/`. `MANIFEST.json` carries per-file evidence and gate status.

	# models/: imatrix-calibrated weights (G30, 2026-10-14)

	Same quant types and byte budget as the shipped files, re-quantized with an importance matrix.
	Additive only: the phone app still fetches `kev-4b-q4km/` at the repo root, and that dir is unchanged.

	\| file \| bytes \| sha256 \| status \|
	\|---\|---\|---\|---\|
	\| `kev-4b-q4km-imat/Kev-4B-Q4_K_M-imat.gguf` \| 2,709,547,712 \| `0535181cf3164a7271a7d3d1752a9ceab673c32bb316a081133ec238c186b51c` \| ship (llama.cpp path); app NOT repointed \|
	\| `kev-4b-iq4xs-imat/Kev-4B-IQ4_XS-imat.gguf` \| 2,514,987,712 \| `8a45e54a03a689368413de3bb6d33c1129ea416dfa7a8aade64c2f8c2c3b47dc` \| published for llama.cpp; native path BLOCKED until gate receipt \|
	\| `qwen38-27b-q4ks-imat-openjev-wide/Qwen3.8-27B-Q4_K_S-imat-openjev-wide.gguf` \| 15,586,316,480 \| `c1e505b2e4266d964cc75e8a801e1c54dfec1bd4205c6ea7072d15c9b907fc0f` \| ship (board, llama-server); promotion into the board store is an owner action \|
	\| `imatrix/kev4b-calib.imatrix.gguf` \| 3,626,496 \| `fe439522f2acc99fd040960f045f9582ccb89fe055d570ca883c61600a2f4737` \| provenance \|
	\| `imatrix/qwen38-27b-calib.imatrix.gguf` \| 13,642,656 \| `bd707218d6275dbcb342b23846776b306e535f431cb76f316349fa451138488c` \| provenance \|

	## Fidelity (llama.cpp `--kl-divergence`, 40×512 ctx, 10,200 tokens, held-out lab prose)
	\| artifact \| ref \| mean KLD \| top-1 % \|
	\|---\|---\|---\|---\|
	\| Kev-4B Q4_K_M plain (shipped) \| Kev BF16 \| 0.04573 \| 88.25 \|
	\| Kev-4B Q4_K_M-imat \| Kev BF16 \| 0.02894 (−37%) \| 89.81 \|
	\| Kev-4B IQ4_XS-imat \| Kev BF16 \| 0.03593 \| 89.22 \|
	\| 27B Q4_K_S plain (shipped) \| 27B Q6_K \| 0.03652 \| 90.84 \|
	\| 27B Q4_K_S-imat \| 27B Q6_K \| 0.02126 (−42%) \| 93.30 \|

	The calibration data is in-domain (disjoint from the eval set), so expect a smaller gain on generic text.
	The decision battery (40 scored questions) shows no regression vs shipped for any of these files.

	IQ4_XS-imat: the native `llb` engine gate is still pending. Use it only with llama.cpp until that gate passes.

	Verify: `sha256sum -c SHA256SUMS` inside `models/`. `MANIFEST.json` carries per-file evidence and gate status.