Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,90 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- en
|
4 |
+
library_name: peft
|
5 |
+
pipeline_tag: text-generation
|
6 |
+
tags:
|
7 |
+
- medical
|
8 |
+
license: llama2
|
9 |
+
datasets:
|
10 |
+
- nmitchko/i2b2-query-data-1.0
|
11 |
+
---
|
12 |
+
|
13 |
+
# i2b2 QueryBuilder - 34b Merged
|
14 |
+
|
15 |
+
<!-- TODO: Add a link here N: DONE-->
|
16 |
+
![Screenshot](https://huggingface.co/nmitchko/i2b2-querybuilder-codellama-34b/resolve/main/Example%20Query.png)
|
17 |
+
|
18 |
+
## Model Description
|
19 |
+
|
20 |
+
This model will generate queries for your i2b2 query builder trained on [this dataset](https://huggingface.co/datasets/nmitchko/i2b2-query-data-1.0) for `10 epochs` . For evaluation use.
|
21 |
+
* Do not use as a final research query builder.
|
22 |
+
* Results may be incorrect or mal-formatted.
|
23 |
+
* The onus of research accuracy is on the researcher, not the AI model.
|
24 |
+
|
25 |
+
## Prompt Format
|
26 |
+
|
27 |
+
If you are using text-generation-webui, you can download the instruction template [i2b2.yaml](https://huggingface.co/nmitchko/i2b2-querybuilder-codellama-34b/resolve/main/i2b2.yaml)
|
28 |
+
|
29 |
+
```md
|
30 |
+
Below is an instruction that describes a task.
|
31 |
+
|
32 |
+
### Instruction:
|
33 |
+
{input}
|
34 |
+
|
35 |
+
### Response:
|
36 |
+
```xml
|
37 |
+
```
|
38 |
+
|
39 |
+
|
40 |
+
### Architecture
|
41 |
+
`nmitchko/i2b2-querybuilder-34b-merged` is a large language model LoRa specifically fine-tuned for generating queries in the [i2b2 query builder](https://community.i2b2.org/wiki/display/webclient/3.+Query+Tool).
|
42 |
+
It is based on [`codellama-34b-hf`](https://huggingface.co/codellama/CodeLlama-34b-hf) at 34 billion parameters.
|
43 |
+
|
44 |
+
The primary goal of this model is to improve research accuracy with the i2b2 tool.
|
45 |
+
It was trained using [LoRA](https://arxiv.org/abs/2106.09685), specifically [QLora Multi GPU](https://github.com/ChrisHayduk/qlora-multi-gpu), to reduce memory footprint.
|
46 |
+
|
47 |
+
See Training Parameters for more info This Lora supports 4-bit and 8-bit modes.
|
48 |
+
|
49 |
+
### Requirements
|
50 |
+
|
51 |
+
```
|
52 |
+
bitsandbytes>=0.41.0
|
53 |
+
peft@main
|
54 |
+
transformers@main
|
55 |
+
```
|
56 |
+
|
57 |
+
Steps to load this model:
|
58 |
+
1. Load base model (codellama-34b-hf) using transformers
|
59 |
+
|
60 |
+
## Training Parameters
|
61 |
+
|
62 |
+
The model was trained for or 10 epochs on [i2b2-query-data-1.0](https://huggingface.co/datasets/nmitchko/i2b2-query-data-1.0)
|
63 |
+
`i2b2-query-data-1.0` contains only tasks and outputs for i2b2 queries xsd schemas.
|
64 |
+
|
65 |
+
|
66 |
+
| Item | Amount | Units |
|
67 |
+
|---------------|--------|-------|
|
68 |
+
| LoRA Rank | 64 | ~ |
|
69 |
+
| LoRA Alpha | 16 | ~ |
|
70 |
+
| Learning Rate | 1e-4 | SI |
|
71 |
+
| Dropout | 5 | % |
|
72 |
+
|
73 |
+
## Training procedure
|
74 |
+
|
75 |
+
|
76 |
+
The following `bitsandbytes` quantization config was used during training:
|
77 |
+
- quant_method: QuantizationMethod.BITS_AND_BYTES
|
78 |
+
- load_in_8bit: False
|
79 |
+
- load_in_4bit: True
|
80 |
+
- llm_int8_threshold: 6.0
|
81 |
+
- llm_int8_skip_modules: None
|
82 |
+
- llm_int8_enable_fp32_cpu_offload: False
|
83 |
+
- llm_int8_has_fp16_weight: False
|
84 |
+
- bnb_4bit_quant_type: nf4
|
85 |
+
- bnb_4bit_use_double_quant: True
|
86 |
+
- bnb_4bit_compute_dtype: bfloat16
|
87 |
+
|
88 |
+
### Framework versions
|
89 |
+
|
90 |
+
- PEFT 0.6.0.dev0
|