GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
tomhu
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
tomhu
Model Tree
Qwen/Qwen3-8B
Input Modalities
Output Modalities
Supported Functionality
| Setting | Value |
|---|---|
| Student | Qwen/Qwen3-8B |
| Teacher | Qwen/Qwen3-14B |
| Objective | forward KL over teacher top-k logits |
| Teacher top-k | 256 |
| Policy-gradient term | disabled |
| Precision | BF16 |
| Learning rate | 3e-6, constant schedule |
| Epochs | 1 |
| Global batch size | 32 |
| Per-GPU micro batch | 1 |
| Gradient accumulation | 32 |
| Rollout samples | 1 |
| Temperature | 1.0 |
| Maximum prompt/response length | 4096 / 4096 tokens |
| Seed | 42 |
| Setting | Value |
|---|---|
| Starting policy | OPD model from stage 1 |
| Training examples | 1,582 mutation-applicable Defects4J training samples |
| Algorithm | GRPO, no KL term in reward |
| Configured epochs | 2 |
| Learning rate | 1e-6 |
| Global prompt batch | 32 |
| Rollouts per prompt | 8 |
| PPO prompt groups / expanded mini batch | 4 / 32 |
The reward is -1 for an invalid generated test. A valid test receives
0.4 * line_coverage + 0.6 * branch_coverage + mutation_score. Mutation
analysis samples up to three eligible failing mutants per rollout, with a
mutation weight of 1.0. Coverage credit is gated by focal-call, assertion, and
local focal-data provenance checks.
The model is intended for research on Java unit-test generation and reinforcement learning for software engineering. Generated tests must be compiled and executed in an isolated environment before use.
| Per-GPU micro batch | 2 |
| Gradient accumulation | 8 |
| Temperature | 1.0 |
| Entropy coefficient | 0.01 |
| Maximum prompt/response length | 4096 / 4096 tokens |
| Data shuffle / seed | enabled / 42 |
| Chat template | enabled; Qwen thinking disabled |