GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Jainamshahhh
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
Jainamshahhh
Model Tree
Qwen/Qwen2.5-1.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Four model generations, every gate pre-registered and reported:
| gen | held-out action agreement | plan-conditioning story |
|---|---|---|
| v1 | 77.1% | plan-DEAF (ΔP 0.07; no-plan ablation cost just 1.8pts) |
| v2 | 69.5% | conditioning 2-3× base; rush↔turtle style separation 0.23 vs teacher 0.60 |
| v3 | 73.5% | teacher-identical style separation (0.602 vs 0.600) — strike-suppression learned |
| v4 (this) | 73.9% | strongest fighter; plan-conditioning regressed (data dilution — see Field Notes) |
The pre-registered ΔP gate FAILED on every generation as registered — then we measured the gate's ceiling and found the teacher itself couldn't pass the direction threshold. Full post-mortems (yardstick calibration, the counterfactual-data fix, the engine design bug a playtest exposed) live in the project Field Notes.
bash
# data (Node, deterministic): npx tsx training/datagen/run.tsmodal run train_bc.py --data-dir /vol/data/duels_v4 --run-name bc_v4 --epochs 2modal run intervention.py::intervention_entrypoint --run-name bc_v4 --data-dir /vol/data/duels_v4Trained on Modal (A100); served for dev via vLLM with structured_outputs.choice
over the action alphabet. The shipped Space runs 100% in-browser per organizer
guidance — this model is the published artifact + the dev/video opponent.
(LoRA adapter repo — see the merged repo for standalone weights.)
# data (Node, deterministic): npx tsx training/datagen/run.tsmodal run train_bc.py --data-dir /vol/data/duels_v4 --run-name bc_v4 --epochs 2modal run intervention.py::intervention_entrypoint --run-name bc_v4 --data-dir /vol/data/duels_v4