Update README.md
Browse files
README.md
CHANGED
|
@@ -11,6 +11,93 @@ tags:
|
|
| 11 |
This model was converted to GGUF format from [`arcee-ai/Arcee-Maestro-7B-Preview`](https://huggingface.co/arcee-ai/Arcee-Maestro-7B-Preview) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space.
|
| 12 |
Refer to the [original model card](https://huggingface.co/arcee-ai/Arcee-Maestro-7B-Preview) for more details on the model.
|
| 13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
## Use with llama.cpp
|
| 15 |
Install llama.cpp through brew (works on Mac and Linux)
|
| 16 |
|
|
|
|
| 11 |
This model was converted to GGUF format from [`arcee-ai/Arcee-Maestro-7B-Preview`](https://huggingface.co/arcee-ai/Arcee-Maestro-7B-Preview) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space.
|
| 12 |
Refer to the [original model card](https://huggingface.co/arcee-ai/Arcee-Maestro-7B-Preview) for more details on the model.
|
| 13 |
|
| 14 |
+
---
|
| 15 |
+
Arcee-Maestro-7B-Preview (7B) is Arcee's first reasoning model trained with reinforment learning. It is based on the Qwen2.5-7B DeepSeek-R1 distillation DeepSeek-R1-Distill-Qwen-7B with further GPRO training. Though this is just a preview of our
|
| 16 |
+
upcoming work, it already shows promising improvements to mathematical
|
| 17 |
+
and coding abilities across a range of tasks.
|
| 18 |
+
|
| 19 |
+
Intended Use Cases
|
| 20 |
+
-
|
| 21 |
+
|
| 22 |
+
Advanced reasoning
|
| 23 |
+
|
| 24 |
+
Mathematics
|
| 25 |
+
|
| 26 |
+
Coding
|
| 27 |
+
|
| 28 |
+
Training & Fine-Tuning
|
| 29 |
+
-
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
Initial Training: Began with DeepSeek-R1-Distill-Qwen-7B
|
| 34 |
+
GRPO:
|
| 35 |
+
Trained on 450,000 verified math problems
|
| 36 |
+
Additional bootstrapped coding examples
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
Performance
|
| 47 |
+
-
|
| 48 |
+
|
| 49 |
+
|
| 50 |
+
|
| 51 |
+
Arcee-Maestro-7B-Preview shows strong performance in mathematics as
|
| 52 |
+
well as coding, competing against even O1 preview, a model far
|
| 53 |
+
surprassing its size.
|
| 54 |
+
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
Limitations
|
| 62 |
+
-
|
| 63 |
+
|
| 64 |
+
|
| 65 |
+
|
| 66 |
+
Context Length: 128k Tokens (may vary depending on the final tokenizer settings and system resources).
|
| 67 |
+
Knowledge Cut-off: Training data may not reflect the latest events or developments beyond June 2024.
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
|
| 71 |
+
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
Ethical Considerations
|
| 76 |
+
-
|
| 77 |
+
|
| 78 |
+
|
| 79 |
+
|
| 80 |
+
Content Generation Risks: Like any language model,
|
| 81 |
+
Arcee-Maestro-7B-Preview can generate potentially harmful or biased
|
| 82 |
+
content if prompted in certain ways.
|
| 83 |
+
|
| 84 |
+
|
| 85 |
+
|
| 86 |
+
|
| 87 |
+
|
| 88 |
+
|
| 89 |
+
|
| 90 |
+
License
|
| 91 |
+
-
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
|
| 95 |
+
Arcee-Maestro-7B-Preview (7B) is released under the Apache-2.0 License.
|
| 96 |
+
You are free to use, modify, and distribute this model in both
|
| 97 |
+
commercial and non-commercial applications, subject to the terms and
|
| 98 |
+
conditions of the license.
|
| 99 |
+
|
| 100 |
+
---
|
| 101 |
## Use with llama.cpp
|
| 102 |
Install llama.cpp through brew (works on Mac and Linux)
|
| 103 |
|