Commit
ยท
5c4f2fd
1
Parent(s):
acf6917
Update README.md to enhance model description and add advanced example for ArXiv ML trends analysis
Browse files
README.md
CHANGED
@@ -32,6 +32,7 @@ That's it! No installation, no setup - just `uv run`.
|
|
32 |
- **Guaranteed valid outputs** using vLLM's guided decoding with outlines
|
33 |
- **Zero-shot classification** with structured generation
|
34 |
- **GPU-optimized** with vLLM's automatic batching for maximum efficiency
|
|
|
35 |
- **Robust text handling** with preprocessing and validation
|
36 |
- **Three prompt styles** for different use cases
|
37 |
- **Automatic progress tracking** and detailed statistics
|
@@ -52,13 +53,15 @@ uv run classify-dataset.py \
|
|
52 |
### Arguments
|
53 |
|
54 |
**Required:**
|
|
|
55 |
- `--input-dataset`: Hugging Face dataset ID (e.g., `stanfordnlp/imdb`, `user/my-dataset`)
|
56 |
- `--column`: Name of the text column to classify
|
57 |
- `--labels`: Comma-separated classification labels (e.g., `"spam,ham"`)
|
58 |
- `--output-dataset`: Where to save the classified dataset
|
59 |
|
60 |
**Optional:**
|
61 |
-
|
|
|
62 |
- `--prompt-style`: Choose from `simple`, `detailed`, or `reasoning` (default: `simple`)
|
63 |
- `--split`: Dataset split to process (default: `train`)
|
64 |
- `--max-samples`: Limit samples for testing
|
@@ -77,6 +80,7 @@ All styles benefit from structured output guarantees - the model can only output
|
|
77 |
## ๐ Examples
|
78 |
|
79 |
### Sentiment Analysis
|
|
|
80 |
```bash
|
81 |
uv run classify-dataset.py \
|
82 |
--input-dataset stanfordnlp/imdb \
|
@@ -86,6 +90,7 @@ uv run classify-dataset.py \
|
|
86 |
```
|
87 |
|
88 |
### Support Ticket Classification
|
|
|
89 |
```bash
|
90 |
uv run classify-dataset.py \
|
91 |
--input-dataset user/support-tickets \
|
@@ -96,6 +101,7 @@ uv run classify-dataset.py \
|
|
96 |
```
|
97 |
|
98 |
### News Categorization
|
|
|
99 |
```bash
|
100 |
uv run classify-dataset.py \
|
101 |
--input-dataset ag_news \
|
@@ -109,30 +115,17 @@ uv run classify-dataset.py \
|
|
109 |
|
110 |
This script is optimized for [Hugging Face Jobs](https://huggingface.co/docs/hub/spaces-gpu-jobs) (requires Pro subscription or Team/Enterprise organization):
|
111 |
|
112 |
-
|
113 |
# Run on L4 GPU with vLLM image
|
114 |
hf jobs uv run \
|
115 |
--flavor l4x1 \
|
116 |
--image vllm/vllm-openai:latest \
|
117 |
-
classify-dataset.py \
|
118 |
--input-dataset stanfordnlp/imdb \
|
119 |
--column text \
|
120 |
--labels "positive,negative" \
|
121 |
--output-dataset user/imdb-classified
|
122 |
|
123 |
-
# Run on A10 GPU with custom model
|
124 |
-
hf jobs uv run \
|
125 |
-
--flavor a10g-large \
|
126 |
-
--image vllm/vllm-openai:latest \
|
127 |
-
classify-dataset.py \
|
128 |
-
--input-dataset user/reviews \
|
129 |
-
--column review_text \
|
130 |
-
--labels "1,2,3,4,5" \
|
131 |
-
--output-dataset user/reviews-rated \
|
132 |
-
--model mistralai/Mistral-7B-Instruct-v0.3 \
|
133 |
-
--prompt-style detailed
|
134 |
-
```
|
135 |
-
|
136 |
### GPU Flavors
|
137 |
- `t4-small`: Budget option for smaller models
|
138 |
- `l4x1`: Good balance for 7B models
|
@@ -144,7 +137,7 @@ hf jobs uv run \
|
|
144 |
|
145 |
### Using Different Models
|
146 |
|
147 |
-
|
148 |
|
149 |
```bash
|
150 |
# Larger model for complex classification
|
@@ -154,7 +147,7 @@ uv run classify-dataset.py \
|
|
154 |
--labels "contract,patent,brief,memo,other" \
|
155 |
--output-dataset user/legal-classified \
|
156 |
--model Qwen/Qwen2.5-7B-Instruct
|
157 |
-
|
158 |
|
159 |
### Large Datasets
|
160 |
|
@@ -170,7 +163,7 @@ uv run classify-dataset.py \
|
|
170 |
|
171 |
## ๐ Performance
|
172 |
|
173 |
-
- **SmolLM3-3B**: ~50-100 texts/second on A10
|
174 |
- **7B models**: ~20-50 texts/second on A10
|
175 |
- vLLM automatically optimizes batching for best throughput
|
176 |
|
@@ -186,32 +179,76 @@ The script loads your dataset, preprocesses texts, classifies each one using gui
|
|
186 |
## ๐ Troubleshooting
|
187 |
|
188 |
### CUDA Not Available
|
|
|
189 |
This script requires a GPU. Run it on:
|
|
|
190 |
- A machine with NVIDIA GPU
|
191 |
- HF Jobs (recommended)
|
192 |
- Cloud GPU instances
|
193 |
|
194 |
### Out of Memory
|
|
|
195 |
- Use a smaller model
|
196 |
- Use a larger GPU (e.g., a100-large)
|
197 |
|
198 |
### Invalid/Skipped Texts
|
|
|
199 |
- Texts shorter than 3 characters are skipped
|
200 |
- Empty or None values are marked as invalid
|
201 |
- Very long texts are truncated to 4000 characters
|
202 |
|
203 |
### Classification Quality
|
|
|
204 |
- With guided decoding, outputs are guaranteed to be valid labels
|
205 |
- For better results, use clear and distinct label names
|
206 |
- Try the `reasoning` prompt style for complex classifications
|
207 |
- Use a larger model for nuanced tasks
|
208 |
|
209 |
### vLLM Version Issues
|
|
|
210 |
If you see `ImportError: cannot import name 'GuidedDecodingParams'`:
|
|
|
211 |
- Your vLLM version is too old (requires >= 0.6.6)
|
212 |
- The script specifies the correct version in its dependencies
|
213 |
- UV should automatically install the correct version
|
214 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
215 |
## ๐ License
|
216 |
|
217 |
-
This script is provided as-is for use with the UV Scripts organization.
|
|
|
32 |
- **Guaranteed valid outputs** using vLLM's guided decoding with outlines
|
33 |
- **Zero-shot classification** with structured generation
|
34 |
- **GPU-optimized** with vLLM's automatic batching for maximum efficiency
|
35 |
+
- **Default model**: HuggingFaceTB/SmolLM3-3B (fast 3B model, easily changeable)
|
36 |
- **Robust text handling** with preprocessing and validation
|
37 |
- **Three prompt styles** for different use cases
|
38 |
- **Automatic progress tracking** and detailed statistics
|
|
|
53 |
### Arguments
|
54 |
|
55 |
**Required:**
|
56 |
+
|
57 |
- `--input-dataset`: Hugging Face dataset ID (e.g., `stanfordnlp/imdb`, `user/my-dataset`)
|
58 |
- `--column`: Name of the text column to classify
|
59 |
- `--labels`: Comma-separated classification labels (e.g., `"spam,ham"`)
|
60 |
- `--output-dataset`: Where to save the classified dataset
|
61 |
|
62 |
**Optional:**
|
63 |
+
|
64 |
+
- `--model`: Model to use (default: **`HuggingFaceTB/SmolLM3-3B`** - a fast 3B parameter model)
|
65 |
- `--prompt-style`: Choose from `simple`, `detailed`, or `reasoning` (default: `simple`)
|
66 |
- `--split`: Dataset split to process (default: `train`)
|
67 |
- `--max-samples`: Limit samples for testing
|
|
|
80 |
## ๐ Examples
|
81 |
|
82 |
### Sentiment Analysis
|
83 |
+
|
84 |
```bash
|
85 |
uv run classify-dataset.py \
|
86 |
--input-dataset stanfordnlp/imdb \
|
|
|
90 |
```
|
91 |
|
92 |
### Support Ticket Classification
|
93 |
+
|
94 |
```bash
|
95 |
uv run classify-dataset.py \
|
96 |
--input-dataset user/support-tickets \
|
|
|
101 |
```
|
102 |
|
103 |
### News Categorization
|
104 |
+
|
105 |
```bash
|
106 |
uv run classify-dataset.py \
|
107 |
--input-dataset ag_news \
|
|
|
115 |
|
116 |
This script is optimized for [Hugging Face Jobs](https://huggingface.co/docs/hub/spaces-gpu-jobs) (requires Pro subscription or Team/Enterprise organization):
|
117 |
|
118 |
+
````bash
|
119 |
# Run on L4 GPU with vLLM image
|
120 |
hf jobs uv run \
|
121 |
--flavor l4x1 \
|
122 |
--image vllm/vllm-openai:latest \
|
123 |
+
https://huggingface.co/datasets/uv-scripts/classification/raw/main/classify-dataset.py \
|
124 |
--input-dataset stanfordnlp/imdb \
|
125 |
--column text \
|
126 |
--labels "positive,negative" \
|
127 |
--output-dataset user/imdb-classified
|
128 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
129 |
### GPU Flavors
|
130 |
- `t4-small`: Budget option for smaller models
|
131 |
- `l4x1`: Good balance for 7B models
|
|
|
137 |
|
138 |
### Using Different Models
|
139 |
|
140 |
+
By default, this script uses **HuggingFaceTB/SmolLM3-3B** - a fast, efficient 3B parameter model that's perfect for most classification tasks. You can easily use any other instruction-tuned model:
|
141 |
|
142 |
```bash
|
143 |
# Larger model for complex classification
|
|
|
147 |
--labels "contract,patent,brief,memo,other" \
|
148 |
--output-dataset user/legal-classified \
|
149 |
--model Qwen/Qwen2.5-7B-Instruct
|
150 |
+
````
|
151 |
|
152 |
### Large Datasets
|
153 |
|
|
|
163 |
|
164 |
## ๐ Performance
|
165 |
|
166 |
+
- **SmolLM3-3B (default)**: ~50-100 texts/second on A10
|
167 |
- **7B models**: ~20-50 texts/second on A10
|
168 |
- vLLM automatically optimizes batching for best throughput
|
169 |
|
|
|
179 |
## ๐ Troubleshooting
|
180 |
|
181 |
### CUDA Not Available
|
182 |
+
|
183 |
This script requires a GPU. Run it on:
|
184 |
+
|
185 |
- A machine with NVIDIA GPU
|
186 |
- HF Jobs (recommended)
|
187 |
- Cloud GPU instances
|
188 |
|
189 |
### Out of Memory
|
190 |
+
|
191 |
- Use a smaller model
|
192 |
- Use a larger GPU (e.g., a100-large)
|
193 |
|
194 |
### Invalid/Skipped Texts
|
195 |
+
|
196 |
- Texts shorter than 3 characters are skipped
|
197 |
- Empty or None values are marked as invalid
|
198 |
- Very long texts are truncated to 4000 characters
|
199 |
|
200 |
### Classification Quality
|
201 |
+
|
202 |
- With guided decoding, outputs are guaranteed to be valid labels
|
203 |
- For better results, use clear and distinct label names
|
204 |
- Try the `reasoning` prompt style for complex classifications
|
205 |
- Use a larger model for nuanced tasks
|
206 |
|
207 |
### vLLM Version Issues
|
208 |
+
|
209 |
If you see `ImportError: cannot import name 'GuidedDecodingParams'`:
|
210 |
+
|
211 |
- Your vLLM version is too old (requires >= 0.6.6)
|
212 |
- The script specifies the correct version in its dependencies
|
213 |
- UV should automatically install the correct version
|
214 |
|
215 |
+
## ๐ฌ Advanced Example: ArXiv ML Trends Analysis
|
216 |
+
|
217 |
+
For a more complex real-world example, we provide scripts to analyze ML research trends from ArXiv papers:
|
218 |
+
|
219 |
+
### Step 1: Prepare the Dataset
|
220 |
+
|
221 |
+
```bash
|
222 |
+
# Filter and prepare ArXiv CS papers from 2024
|
223 |
+
uv run prepare_arxiv_2024.py
|
224 |
+
```
|
225 |
+
|
226 |
+
This creates a filtered dataset of CS papers with combined title+abstract text.
|
227 |
+
|
228 |
+
### Step 2: Run Classification with Python API
|
229 |
+
|
230 |
+
```bash
|
231 |
+
# Use HF Jobs Python API to classify papers
|
232 |
+
uv run run_arxiv_classification.py
|
233 |
+
```
|
234 |
+
|
235 |
+
This script demonstrates:
|
236 |
+
|
237 |
+
- Using `run_uv_job()` from the Python API
|
238 |
+
- Classifying into modern ML trends (reasoning, agents, multimodal, robotics, etc.)
|
239 |
+
- Handling authentication and job monitoring
|
240 |
+
|
241 |
+
The classification categories include:
|
242 |
+
|
243 |
+
- `reasoning_systems`: Chain-of-thought, reasoning, problem solving
|
244 |
+
- `agents_autonomous`: Agents, tool use, autonomous systems
|
245 |
+
- `multimodal_models`: Vision-language, audio, multi-modal
|
246 |
+
- `robotics_embodied`: Robotics, embodied AI, manipulation
|
247 |
+
- `efficient_inference`: Quantization, distillation, edge deployment
|
248 |
+
- `alignment_safety`: RLHF, alignment, safety, interpretability
|
249 |
+
- `generative_models`: Diffusion, generation, synthesis
|
250 |
+
- `foundational_other`: Other foundational ML/AI research
|
251 |
+
|
252 |
## ๐ License
|
253 |
|
254 |
+
This script is provided as-is for use with the UV Scripts organization.
|