GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
chiuratto-AIgourakis
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
chiuratto-AIgourakis
Model Tree
Qwen/Qwen2.5-Coder-1.5B
Input Modalities
Output Modalities
Supported Functionality
Qwen/Qwen2.5-Coder-1.5B
| Field | Value |
|---|---|
| Dataset branch commit | aa5f9218e |
| Training rows requested | 4,800 |
| Training rows used | 4,769 |
| Optimiser steps | 120 |
| Repair-pair rows | 480 |
| Run-expected rows | 435 |
| Prompt style | sounio-compact-completion |
| LoRA rank | 16 |
| LoRA alpha | 32 |
Held-out split: datasets/hf_examples/instruction_pairs_eval.jsonl, 200 rows.
| Model/objective | Compile pass | Compile + contract-clean | Exact runtime pass |
|---|---|---|---|
| Base model, zero training | 111/200 (55.5%) | not recorded | 25/65 (38.5%) |
| Adapter candidate | 192/200 (96.0%) | 192/200 (96.0%) | 58/65 (89.2%) |
| Adapter + structural repair layer | 199/200 (99.5%) | 198/200 (99.0%) | 59/65 (90.8%) |
The 98%+ result uses the saved adapter plus the deterministic structural repair
layer in the Sounio repository. The repair layer catches outputs that are still
outside the complete-file objective, such as Rust-shaped String/Vec
surfaces, invalid synthetic declarations, algebra blocks used as structs, and
one observed scalar typo. It does not retrain model weights.
adapter_model.safetensorsadapter_config.jsontrain_summary.jsonlora_smoke_eval_report.jsonadapter_eval_summary.jsonadapter_eval_report.jsonADAPTER_CANDIDATE_QWEN15.mdPrimary adapter training run:
text
/orangefs/training/sounio/hf-examples/lora-smoke/sounio-qwen15-lora-full200-surgical-repair-20260522T021535/results98%+ adapter-only evaluation run:
text
/orangefs/training/sounio/hf-examples/lora-smoke/sounio-qwen15-adapter-eval-98plus-20260522T083913/resultsKey checksum:
text
9d8540e218439bb28026115fbe8d4be5fe18162a0a1c07474d9850579ee36c4a adapter_model.safetensorsUse this adapter to generate or repair small Sounio examples under the
sounio-compact-completion prompt format. Outputs should still be validated
with souc check and, when applicable, runtime checks.