deepscaler:latest
Upstream reference: DeepScaleR 1.5B Preview
Publisher: Agentica
DeepScaleR is a mathematics-oriented language-model research release post-trained from DeepSeek-R1-Distill-Qwen-1.5B using distributed reinforcement learning.
Publisher-described context, separate from Sigma5C observations and recorded catalog statistics.
Published upstream specifications
- Published size label
- 1.5B Source
- Upstream derivative license label
- MIT Source
- Documented training method
- GRPO with iterative context lengthening Source
Publisher-described uses
- Mathematical text generation and reinforcement-learning research. Source
Reference limitations
- Provider evaluation results are specific to its research conditions and are not Sigma5C observations. Source
A recorded tag and an upstream card association do not verify the deployed checkpoint, quantization, configured context or current hosted availability.
Publisher sources
- DeepScaleR 1.5B Preview official documentation — Agentica. Checked .
Links
Recorded references do not establish downloadable weights, licensing or current availability.
1.5B
parameters
F16
quantization
qwen2
architecture
131072
context
mit
license
2025-02-10
released
Where deepscaler:latest comes from
Company / lab
Agentica
See every model and family from this maker.
Model family
deepscaler
Other versions and sizes in this line.
completionthinking