Palmyra-mini-thinking-b
Release in the Palmyra family · version palmyra-mini-thinking-b
A small reasoning model in Writer's Palmyra-mini family, released in September 2025. Writer built it on NVIDIA's OpenReasoning-Nemotron-1.5B and applied reinforcement-learning fine-tuning; the model card lists a 131,072-token context window.12
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Model hub: GGUF build (external site: huggingface.co)
- Model hub: MLX build (external site: huggingface.co)
- Release notes: Palmyra-mini announcement (Hugging Face blog) (external site: huggingface.co)
Availability and license
Overall availability
Downloadable from Hugging Face without gating.1
Availability is separate from permission: read the license before using or redistributing.
Writer's card lists Apache 2.0 and does not mention the base model's license. NVIDIA's card for the base model, OpenReasoning-Nemotron-1.5B, says its use is governed by the Creative Commons Attribution 4.0 license (CC-BY-4.0) and links the Apache 2.0 license from a Qwen repository as additional information. The two licenses differ, and Writer's card does not explain how the base model's CC-BY-4.0 terms apply.14
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | Writer also publishes GGUF and MLX conversions.13 |
| Inference codeIs code for running the model published? | Public | The card documents inference with Hugging Face Transformers and serving with vLLM.1 |
| Training codeIs the code used to train the model published? | Unknown | Not assessed. |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Unknown | Neither the model card nor Writer's announcement names the fine-tuning data.12 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | Writer's announcement says the thinking models were trained with a chain-of-thought approach and that this model received reinforcement-learning fine-tuning; no further detail is given.2 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The card publishes benchmark scores and names the tools used (lm_eval, lighteval, and nemoskills) without publishing evaluation configurations.1 |
What it is useful for
Organization context
Provenance and derivatives
Fine-tuned by Writer from NVIDIA's OpenReasoning-Nemotron-1.5B. NVIDIA's card says that model was developed from Qwen2.5-1.5B (Alibaba Cloud's Qwen family) and post-trained on responses generated by DeepSeek-R1-0528.147
- Derived from: OpenReasoning-Nemotron-1.5B (external site: huggingface.co) — Direct base model (NVIDIA).
- Derived from: Qwen2.5-1.5B (external site: huggingface.co) — Base of OpenReasoning-Nemotron-1.5B.
Other releases in the Palmyra family
- Palmyra-miniModel-disclosure tier (USASI rubric v0.1): Open-weight
U.S. eligibility
Eligible · basis: U.S. headquarters
Published by Writer, Inc., which is headquartered in San Francisco per its company page and platform services agreement. Its base model is NVIDIA's OpenReasoning-Nemotron-1.5B, which NVIDIA's card describes as a derivative of Qwen2.5-1.5B from Alibaba Cloud's Qwen family; eligibility here rests on Writer as the developer of this fine-tune.15647
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.