GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
pat-jj
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
pat-jj
Model Tree
Base
openai/gpt-oss-20b
Fine-tuned
this model
Input Modalities
Text
Output Modalities
Text
Supported Functionality
Dedicated Endpoints
Container
safetensors shards.
Code: https://github.com/pat-jj/harness-1
Paper: https://arxiv.org/abs/2606.02373
Training data and corpora: https://huggingface.co/datasets/pat-jj/harness-1-train-data
Tinker inference example: https://github.com/pat-jj/harness-1/blob/main/inference/tinker_inference.md
vLLM inference example: https://github.com/pat-jj/harness-1/blob/main/inference/vllm_h100_browsecompplus.md
This repository contains the full merged Harness-1 release checkpoint. The model
is merged into the openai/gpt-oss-20b base model and saved as standard Hugging
Face safetensors shards.
The Harness-1 training data and retrieval corpora are published separately as
pat-jj/harness-1-train-data.
It contains one train split with a stage column:
sft: 899 raw GPT-5.4-generated v8d SFT trajectories produced by generate_sft_ultra_0417.py.rl: 3,453 SEC training-split query records used for RL (TRAIN_DATASETS=sec, RL_QUERY_SPLIT=train).The same dataset repo also includes retrieval corpora under corpora/:
corpora/browsecompplus/test: 1,144,886 chunks.corpora/web/train: 219,388 chunks.corpora/web/test: 54,735 chunks.corpora/patents/train: 104,842 chunks.corpora/patents/test: 35,551 chunks.corpora/sec/train: 2,115,106 chunks.python
from datasets import load_datasetds = load_dataset("pat-jj/harness-1-train-data", split="train")sft = ds.filter(lambda row: row["stage"] == "sft")rl = ds.filter(lambda row: row["stage"] == "rl")sec_corpus = load_dataset("parquet",data_files="hf://datasets/pat-jj/harness-1-train-data/corpora/sec/train/*.parquet",split="train",)