Instructions to use sixstringzen/Hemmingway-1.5-oQ8e-mtp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use sixstringzen/Hemmingway-1.5-oQ8e-mtp with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("sixstringzen/Hemmingway-1.5-oQ8e-mtp")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use sixstringzen/Hemmingway-1.5-oQ8e-mtp with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "sixstringzen/Hemmingway-1.5-oQ8e-mtp"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "sixstringzen/Hemmingway-1.5-oQ8e-mtp"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use sixstringzen/Hemmingway-1.5-oQ8e-mtp with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "sixstringzen/Hemmingway-1.5-oQ8e-mtp"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "sixstringzen/Hemmingway-1.5-oQ8e-mtp"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "sixstringzen/Hemmingway-1.5-oQ8e-mtp",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use sixstringzen/Hemmingway-1.5-oQ8e-mtp with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "sixstringzen/Hemmingway-1.5-oQ8e-mtp"

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 sixstringzen/Hemmingway-1.5-oQ8e-mtp

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use sixstringzen/Hemmingway-1.5-oQ8e-mtp with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "sixstringzen/Hemmingway-1.5-oQ8e-mtp"

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 "sixstringzen/Hemmingway-1.5-oQ8e-mtp" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

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

Hemmingway-1.5 oQ8e with MTP

This repository contains an enhanced oQ8e quantization of HemmingwayAI/Hemmingway-1.5 for MLX and oMLX on Apple silicon. The conversion preserves the model's multi-token prediction (MTP) tensors.

Hemmingway AI developed and published the source model. sixstringzen performed this conversion and published the converted weights and quantization report. Training details, intended use, and upstream limitations are documented in the source model card.

Quantization set

This repository is part of the Hemmingway-1.5 oMLX oQe Quantizations: Apple Silicon collection. All published builds use the same source revision, matching imatrix calibration, bfloat16 for unquantized tensors, and preserved MTP tensors.

Build	Artifact size before publication metadata
oQ2e	10.18 GiB
oQ3e	12.53 GiB
oQ3.5e	13.06 GiB
oQ4e	15.48 GiB
oQ6e	21.39 GiB
oQ8e	27.10 GiB

Quantization details

Item	Value
Source model	`HemmingwayAI/Hemmingway-1.5`
Source revision	`bea9b7a11d670052cd60dfbb55d76a9cbe22d3f5`
Quantizer	oMLX `0.7.3`
Method	Enhanced oQ8e mixed-precision affine quantization
Base precision	8-bit for quantized tensors; protected tensors remain unquantized
Group sizes	64; tensor overrides are in `config.json`
Non-quantized dtype	`bfloat16`
Calibration dataset	`oqe_code_multilingual`
Calibration shape	128 samples at 512 tokens
Imatrix entries	504
Imatrix cache	Reused from the matching oQ4e run
MTP tensors	29 preserved tensors
Artifact size before publication metadata	29,100,362,362 bytes (27.10 GiB)

The included oq_imatrix_report.json records calibration settings, tensor coverage, and fallbacks. It reports one missing imatrix entry for language_model.lm_head and no shape mismatches. Strict imatrix coverage was disabled, allowing the standard oQ fallback for that tensor. The report's local cache path is represented as a relative path for publication.

Compatibility

The converted artifact uses qwen3_5 and standard MLX safetensors. It was produced by oMLX 0.7.3. Other runtime versions and MTP-assisted decoding have not been verified for this release.

Use with oMLX

Download sixstringzen/Hemmingway-1.5-oQ8e-mtp from the oMLX model browser, then load it as an LLM. Enable Lightning MTP separately if your runtime supports it and you want to evaluate MTP decoding.

The upstream model card recommends temperature 1.0. It supports thinking off and the low, medium, and xhigh reasoning settings; consult the upstream card and your runtime for how to select them. Token budgets must cover thinking and the final reply.

Verification

On 2026-11-05, all 6 safetensors shards passed header and file-size checks. The index matched all 1,876 stored tensors, including 29 MTP tensors, with no missing shards or tensor-to-shard mismatches.

This is structural artifact verification. It confirms file and index consistency; inference behavior and quality require separate evaluation.

Quality evaluation

Local generation, numerical fidelity, writing quality, speed, and MTP acceptance have not been benchmarked for these Hemmingway-1.5 builds. Results from the earlier Hemmingway-1 study do not establish results for this release.

Limitations

Quantization can affect word choice, coherence, and instruction following. Calibration used oMLX's built-in dataset rather than a prose-specific dataset. The source model's reported AI-detector results have not been independently reproduced for these quants.

License

Hemmingway-1.5 is distributed under CC BY-NC 4.0. This quantized derivative carries the same non-commercial license and credits Hemmingway AI. See the included LICENSE, copied from the source model, and the upstream model card for commercial licensing information.

The conversion modifies the original weights through mixed-precision quantization and preserves the MTP tensors. It is an unofficial conversion by sixstringzen.