Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,81 @@
|
|
| 1 |
---
|
| 2 |
license: openmdw-1.1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: openmdw-1.1
|
| 3 |
+
base_model:
|
| 4 |
+
- unsloth/Phi-4-mini-reasoning-bnb-4bit
|
| 5 |
+
pipeline_tag: text-generation
|
| 6 |
+
tags:
|
| 7 |
+
- reasoning
|
| 8 |
+
- max
|
| 9 |
+
- compact
|
| 10 |
+
- accessible
|
| 11 |
---
|
| 12 |
+
|
| 13 |
+
<p>
|
| 14 |
+
<img src="aksern.svg" alt="aksern icon" width="112">
|
| 15 |
+
</p>
|
| 16 |
+
|
| 17 |
+
# resolva-1.0-r4b
|
| 18 |
+
|
| 19 |
+
**resolva-1.0-r4b** is a 4B-parameter language model based on **Phi-4-mini-reasoning**, designed to prioritize reasoning effort and task resolution.
|
| 20 |
+
|
| 21 |
+
it is the first model in the resolva series, built around the idea of spending more effort on solving tasks rather than simply generating immediate answers.
|
| 22 |
+
|
| 23 |
+
## model overview
|
| 24 |
+
|
| 25 |
+
- **model name:** resolva-1.0-r4b
|
| 26 |
+
- **series:** resolva
|
| 27 |
+
- **version:** 1.0-r4b
|
| 28 |
+
- **architecture:** Phi-based language model
|
| 29 |
+
- **base model:** Phi-4-mini-reasoning
|
| 30 |
+
- **parameters:** 4B
|
| 31 |
+
- **license:** openmdw-1.1
|
| 32 |
+
- **primary focus:** reasoning and task resolution
|
| 33 |
+
|
| 34 |
+
## key goals
|
| 35 |
+
|
| 36 |
+
- **reasoning-first design** - prioritizes deliberate reasoning when approaching tasks.
|
| 37 |
+
- **task resolution** - designed to work through problems and produce solutions.
|
| 38 |
+
- **efficient scale** - targets a relatively compact 4B-parameter footprint.
|
| 39 |
+
- **foundation for the resolva series** - establishes the starting point for future models and experiments.
|
| 40 |
+
|
| 41 |
+
## intended use
|
| 42 |
+
|
| 43 |
+
resolva-1.0-r4b is intended for tasks that benefit from reasoning, structured problem-solving, and multi-step analysis.
|
| 44 |
+
|
| 45 |
+
potential use cases include:
|
| 46 |
+
- logical reasoning and problem-solving
|
| 47 |
+
- programming and code-related tasks
|
| 48 |
+
- mathematical problems
|
| 49 |
+
- instruction following
|
| 50 |
+
- analytical text generation
|
| 51 |
+
|
| 52 |
+
actual performance may vary depending on the task, inference configuration, and deployment environment.
|
| 53 |
+
|
| 54 |
+
## model details
|
| 55 |
+
|
| 56 |
+
resolva-1.0-r4b is based on Phi-4-mini-reasoning and aims to build on its reasoning-oriented capabilities. it is the first release in the resolva family, providing a foundation for further experimentation with reasoning-focused language models.
|
| 57 |
+
|
| 58 |
+
the `r4b` designation refers to the model's 4B-parameter scale.
|
| 59 |
+
|
| 60 |
+
## limitations
|
| 61 |
+
|
| 62 |
+
- reasoning quality and reliability have not been established across all task categories.
|
| 63 |
+
- the model may generate incorrect or unsupported information.
|
| 64 |
+
- outputs should be evaluated before being used in critical applications.
|
| 65 |
+
- benchmark results and comparative performance should be added once reliable evaluations are available.
|
| 66 |
+
|
| 67 |
+
## usage
|
| 68 |
+
|
| 69 |
+
load the model using the inference framework compatible with its architecture and exported format.
|
| 70 |
+
|
| 71 |
+
refer to the repository files for model weights, tokenizer configuration, and any additional inference instructions.
|
| 72 |
+
|
| 73 |
+
## model series
|
| 74 |
+
|
| 75 |
+
resolva-1.0-r4b is the first model in the resolva series.
|
| 76 |
+
|
| 77 |
+
future releases may explore improvements in reasoning quality, task-solving reliability, and overall model efficiency.
|
| 78 |
+
|
| 79 |
+
---
|
| 80 |
+
|
| 81 |
+
*resolva - focused on reasoning, built for resolution.*
|