GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
menik1126
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
menik1126
Model Tree
Qwen/Qwen2.5-3B
Input Modalities
Output Modalities
Supported Functionality
This is the actor-only inference checkpoint from a 200-step Web Q&A run. The student started from Qwen2.5-3B Base, revision 3aab1f1954e9cc14eb9509a215f9e5ca08227a9b. Query rejection used threshold 5 and sunhaonlp/Simulation_LLM_google_7B_V2 as teacher; think distillation was disabled. Training used batch size 64, mini-batch size 32, learning rate 1e-6, and four AMD GPUs. The run is logged at W&B.
The three model shards, tokenizer and configuration are included. All three uploaded shard hashes match the training output. A single-prompt GPU load-and-generate smoke test passed. This is not a held-out benchmark evaluation; no benchmark score is claimed here. The checkpoint does not include optimizer state and cannot by itself resume training.
See the OVD model collection for other variants. This Base-model pure-Query run is distinct from the Qwen2.5-3B-Instruct Query+CoT paper checkpoint.