Update README.md
Browse files
README.md
CHANGED
|
@@ -59,3 +59,46 @@ hf_pipeline = HuggingFacePipeline(pipeline=generate_code)
|
|
| 59 |
llm_chain = LLMChain(llm=hf_pipeline, prompt=prompt)
|
| 60 |
|
| 61 |
print(llm_chain.predict(instruction="Write a Python function to check if a number is prime."))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
llm_chain = LLMChain(llm=hf_pipeline, prompt=prompt)
|
| 60 |
|
| 61 |
print(llm_chain.predict(instruction="Write a Python function to check if a number is prime."))
|
| 62 |
+
|
| 63 |
+
Known Limitations
|
| 64 |
+
|
| 65 |
+
While Mirror provides high-quality code suggestions, debugging assistance, and structured programming responses, it has the following limitations:
|
| 66 |
+
|
| 67 |
+
General conversation abilities are limited due to its specialization in coding-related tasks.
|
| 68 |
+
Mathematical reasoning and logical inference may be weaker than models designed for general problem-solving.
|
| 69 |
+
Complex multi-step reasoning in natural language might require fine-tuning on additional dialogue datasets.
|
| 70 |
+
Dataset Limitations
|
| 71 |
+
|
| 72 |
+
Mirror is fine-tuned on the GPT CodeFeedback dataset, which primarily focuses on code optimization and structured feedback. While it provides strong performance for technical queries, it may:
|
| 73 |
+
|
| 74 |
+
Reflect biases inherent in publicly available programming datasets.
|
| 75 |
+
Have limited knowledge of recent programming frameworks or libraries that emerged after its last fine-tuning session.
|
| 76 |
+
Exhibit hallucinations in open-ended prompts that lack specific instructions.
|
| 77 |
+
Future Development
|
| 78 |
+
|
| 79 |
+
Enhancing conversational abilities by fine-tuning on instruction-heavy dialogue datasets (e.g., OpenAssistant, Dolly).
|
| 80 |
+
Improving reasoning and debugging capabilities using reinforcement learning from developer interactions.
|
| 81 |
+
Reducing hallucinations in long-form responses through dataset refinements.
|
| 82 |
+
License
|
| 83 |
+
|
| 84 |
+
Mirror is released under the Apache License 2.0 and CC-BY-SA 4.0, allowing for both commercial and research usage.
|
| 85 |
+
|
| 86 |
+
Option 1: Apache License 2.0
|
| 87 |
+
Mirror is licensed under the Apache License, Version 2.0 (the "License");
|
| 88 |
+
you may not use this model except in compliance with the License.
|
| 89 |
+
You may obtain a copy of the License at:
|
| 90 |
+
|
| 91 |
+
📄 Apache 2.0 License
|
| 92 |
+
|
| 93 |
+
Unless required by applicable law or agreed to in writing, software
|
| 94 |
+
distributed under the License is distributed on an "AS IS" BASIS,
|
| 95 |
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
| 96 |
+
|
| 97 |
+
Option 2: Creative Commons Attribution-ShareAlike 4.0 (CC-BY-SA 4.0)
|
| 98 |
+
This model's outputs (such as generated text) and non-code content are licensed under CC-BY-SA 4.0.
|
| 99 |
+
|
| 100 |
+
Under this license:
|
| 101 |
+
|
| 102 |
+
You must give credit when using or sharing outputs.
|
| 103 |
+
You must share modifications under the same license.
|
| 104 |
+
📄 CC-BY-SA 4.0 License
|