Downloads · 30 days
0
qwen2p5-0p5b-traj
arkilpatel/qwen2p5-0p5b-traj
qwen2p5-0p5b-traj is a machine learning model from arkilpatel. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
A dense trajectory of intermediate checkpoints from a single fine-tuning run of Qwen2.5-0.5B on mathematical reasoning.
Access
Public
Updated Aug 25, 2026
Repo size
181 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors175 GB · 96%
At a glance
- License
- apache-2.0
- Access
- Public
- Created
- Aug 25, 2026
- Updated
- Aug 25, 2026
- SHA
- 23f8680d
Base models
Model card
From the Hugging Face model README
Qwen2.5-0.5B fine-tuning checkpoint trajectory
A dense trajectory of intermediate checkpoints from a single fine-tuning run of
Qwen2.5-0.5B on mathematical reasoning.
Contents
| path | what |
|---|---|
step-0025/ … step-4334/ | 177 checkpoints, bf16, inference-ready |
final-full/ | the final checkpoint including optimizer state |
results/ | evaluation outputs |
Checkpoints are saved every 25 optimizer steps across 4 data epochs, so
step-NNNN is the cumulative global step and directory order is training order.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "arkilpatel/qwen2p5-0p5b-traj"
model = AutoModelForCausalLM.from_pretrained(repo, subfolder="step-4334")
tok = AutoTokenizer.from_pretrained(repo, subfolder="step-4334")
Any step-NNNN works as the subfolder.
Notes
- Weights are bfloat16. Training used fp32 master weights; these copies are
cast for inference and are not intended for resuming training.
final-full/retains fp32 weights plus optimizer state and is the only checkpoint from which training could be continued.
- License
- apache-2.0
- Created
- Aug 25, 2026
- Updated
- Aug 25, 2026