julioc-p commited on
Commit
8064dba
·
verified ·
1 Parent(s): 6e0a48b

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +197 -157
README.md CHANGED
@@ -1,202 +1,242 @@
1
  ---
2
  base_model: occiglot/occiglot-7b-eu5
3
  library_name: peft
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4
  ---
5
 
6
- # Model Card for Model ID
7
-
8
- <!-- Provide a quick summary of what the model is/does. -->
9
-
10
-
11
 
12
  ## Model Details
13
 
14
  ### Model Description
15
 
16
- <!-- Provide a longer summary of what this model is. -->
17
-
18
-
19
-
20
- - **Developed by:** [More Information Needed]
21
- - **Funded by [optional]:** [More Information Needed]
22
- - **Shared by [optional]:** [More Information Needed]
23
- - **Model type:** [More Information Needed]
24
- - **Language(s) (NLP):** [More Information Needed]
25
- - **License:** [More Information Needed]
26
- - **Finetuned from model [optional]:** [More Information Needed]
27
-
28
- ### Model Sources [optional]
29
-
30
- <!-- Provide the basic links for the model. -->
31
-
32
- - **Repository:** [More Information Needed]
33
- - **Paper [optional]:** [More Information Needed]
34
- - **Demo [optional]:** [More Information Needed]
35
-
36
- ## Uses
37
-
38
- <!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
39
-
40
- ### Direct Use
41
-
42
- <!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
43
-
44
- [More Information Needed]
45
-
46
- ### Downstream Use [optional]
47
-
48
- <!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
49
-
50
- [More Information Needed]
51
-
52
- ### Out-of-Scope Use
53
 
54
- <!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
55
-
56
- [More Information Needed]
 
 
 
57
 
58
  ## Bias, Risks, and Limitations
59
 
60
- <!-- This section is meant to convey both technical and sociotechnical limitations. -->
61
-
62
- [More Information Needed]
63
-
64
- ### Recommendations
65
-
66
- <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
67
-
68
- Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
69
 
70
  ## How to Get Started with the Model
71
 
72
- Use the code below to get started with the model.
73
-
74
- [More Information Needed]
75
-
76
- ## Training Details
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
77
 
78
  ### Training Data
79
 
80
- <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
81
-
82
- [More Information Needed]
83
-
84
- ### Training Procedure
85
-
86
- <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
87
-
88
- #### Preprocessing [optional]
89
-
90
- [More Information Needed]
91
-
92
 
93
  #### Training Hyperparameters
94
 
95
- - **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
96
-
97
- #### Speeds, Sizes, Times [optional]
98
-
99
- <!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
100
-
101
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
102
 
103
  ## Evaluation
104
 
105
- <!-- This section describes the evaluation protocols and provides the results. -->
106
-
107
  ### Testing Data, Factors & Metrics
108
 
109
  #### Testing Data
110
-
111
- <!-- This should link to a Dataset Card if possible. -->
112
-
113
- [More Information Needed]
114
-
115
- #### Factors
116
-
117
- <!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
118
-
119
- [More Information Needed]
120
 
121
  #### Metrics
122
-
123
- <!-- These are the evaluation metrics being used, ideally with a description of why. -->
124
-
125
- [More Information Needed]
126
 
127
  ### Results
128
 
129
- [More Information Needed]
130
-
131
- #### Summary
132
-
133
-
134
-
135
- ## Model Examination [optional]
136
-
137
- <!-- Relevant interpretability work for the model goes here -->
138
-
139
- [More Information Needed]
 
 
 
 
 
 
140
 
141
  ## Environmental Impact
 
 
 
 
 
142
 
143
- <!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
144
-
145
- Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
146
-
147
- - **Hardware Type:** [More Information Needed]
148
- - **Hours used:** [More Information Needed]
149
- - **Cloud Provider:** [More Information Needed]
150
- - **Compute Region:** [More Information Needed]
151
- - **Carbon Emitted:** [More Information Needed]
152
-
153
- ## Technical Specifications [optional]
154
-
155
- ### Model Architecture and Objective
156
-
157
- [More Information Needed]
158
 
159
  ### Compute Infrastructure
160
 
161
- [More Information Needed]
162
-
163
  #### Hardware
164
-
165
- [More Information Needed]
166
 
167
  #### Software
 
 
168
 
169
- [More Information Needed]
170
-
171
- ## Citation [optional]
172
-
173
- <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
174
-
175
- **BibTeX:**
176
-
177
- [More Information Needed]
178
-
179
- **APA:**
180
-
181
- [More Information Needed]
182
-
183
- ## Glossary [optional]
184
-
185
- <!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
186
-
187
- [More Information Needed]
188
-
189
- ## More Information [optional]
190
-
191
- [More Information Needed]
192
-
193
- ## Model Card Authors [optional]
194
-
195
- [More Information Needed]
196
-
197
- ## Model Card Contact
198
-
199
- [More Information Needed]
200
  ### Framework versions
201
-
202
- - PEFT 0.15.2
 
 
 
 
1
  ---
2
  base_model: occiglot/occiglot-7b-eu5
3
  library_name: peft
4
+ license: mit
5
+ datasets:
6
+ - julioc-p/Question-Sparql
7
+ language:
8
+ - en
9
+ metrics:
10
+ - f1
11
+ - precision
12
+ - recall
13
+ tags:
14
+ - code
15
+ - text-to-sparql
16
+ - sparql
17
+ - wikidata
18
+ - qlora
19
  ---
20
 
21
+ This model is a fine-tuned version of `occiglot/occiglot-7b-eu5` for generating SPARQL queries from English natural language questions, specifically targeting the Wikidata knowledge graph.
 
 
 
 
22
 
23
  ## Model Details
24
 
25
  ### Model Description
26
 
27
+ It was fine-tuned using QLoRA with 4-bit quantization. It takes an English natural language question and corresponding entity/relationship context as input and aims to produce a SPARQL query for Wikidata. This model is part of experiments investigating continual multilingual pre-training.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
28
 
29
+ - **Developed by:** Julio Cesar Perez Duran
30
+ - **Funded by :** DFKI
31
+ - **Model type:** Decoder-only Transformer-based language model
32
+ - **Language(s) (NLP):** en (English)
33
+ - **License:** mit
34
+ - **Finetuned from model:** `occiglot/occiglot-7b-eu5`
35
 
36
  ## Bias, Risks, and Limitations
37
 
38
+ - **Context Reliant:** Performance heavily relies on the provided entity/relationship context mappings.
39
+ - **Output Format:** generates extraneous text after the SPARQL query, requiring post-processing (extraction of content within ` ```sparql ... ``` ` delimiters).
 
 
 
 
 
 
 
40
 
41
  ## How to Get Started with the Model
42
 
43
+ The following Python script provides an example of how to load the model and tokenizer to generate a SPARQL query.
44
+
45
+ ```python
46
+ import torch
47
+ from transformers import AutoTokenizer, BitsAndBytesConfig
48
+ from peft import AutoPeftModelForCausalLM
49
+ import re
50
+ import json
51
+
52
+ model_id = "julioc-p/occiglot_txt_sparql_en_v2"
53
+
54
+ # Configuration for 4-bit quantization
55
+ bnb_config = BitsAndBytesConfig(
56
+ load_in_4bit=True,
57
+ bnb_4bit_quant_type="nf4",
58
+ bnb_4bit_compute_dtype=torch.float16,
59
+ bnb_4bit_use_double_quant=True,
60
+ )
61
+
62
+
63
+ model = AutoPeftModelForCausalLM.from_pretrained(
64
+ model_id,
65
+ quantization_config=bnb_config,
66
+ device_map="auto"
67
+ )
68
+ tokenizer = AutoTokenizer.from_pretrained(model_id)
69
+
70
+ if tokenizer.pad_token is None:
71
+ tokenizer.pad_token = tokenizer.eos_token
72
+ model.config.pad_token_id = tokenizer.pad_token_id
73
+
74
+ # SPARQL extraction function
75
+ def extract_sparql(text):
76
+ code_block_match = re.search(r"```(?:sparql)?\s*(.*?)\s*```", text, re.DOTALL | re.IGNORECASE)
77
+ if code_block_match:
78
+ text_to_search = code_block_match.group(1)
79
+ else:
80
+ text_to_search = text
81
+
82
+ match = re.search(r"(SELECT|ASK|CONSTRUCT|DESCRIBE).*?\}", text_to_search, re.DOTALL | re.IGNORECASE)
83
+ if match:
84
+ return match.group(0).strip()
85
+ return ""
86
+
87
+ # --- Example usage ---
88
+ question = "Who was Barnard College's American female employee?"
89
+ example_context_json_str = '''
90
+ {
91
+ "entities": {
92
+ "Barnard College": "Q167733",
93
+ "American": "Q30",
94
+ "female": "Q6581072",
95
+ "employee": "Q5"
96
+ },
97
+ "relationships": {
98
+ "instance of": "P31",
99
+ "employer": "P108",
100
+ "gender": "P21",
101
+ "country of citizenship": "P27"
102
+ }
103
+ }
104
+ '''
105
+ # System prompt template for v2 models (English)
106
+ system_message_template = """You are an expert text to SparQL query translator. Users will ask you questions in English and you will generate a SparQL query based on the provided context encloses in ```sparql <respose_query>```.
107
+ CONTEXT:
108
+ {context}"""
109
+
110
+ # Format the system message with the actual context
111
+ formatted_system_message = system_message_template.format(context=example_context_json_str)
112
+
113
+ chat_template = [
114
+ {"role": "system", "content": formatted_system_message},
115
+ {"role": "user", "content": question},
116
+ ]
117
+
118
+ inputs = tokenizer.apply_chat_template(
119
+ chat_template,
120
+ tokenize=True,
121
+ add_generation_prompt=True,
122
+ return_tensors="pt"
123
+ ).to(model.device)
124
+
125
+ # Generate the output
126
+ with torch.no_grad():
127
+ outputs = model.generate(
128
+ input_ids=inputs.input_ids,
129
+ attention_mask=inputs.attention_mask,
130
+ max_new_tokens=512,
131
+ do_sample=True,
132
+ temperature=0.7,
133
+ top_p=0.9,
134
+ pad_token_id=tokenizer.pad_token_id
135
+ )
136
+
137
+ # Decode only the generated part
138
+ generated_text_full = tokenizer.decode(outputs[0], skip_special_tokens=True)
139
+ # Extract only assistant's response
140
+ assistant_response_part = generated_text_full.split("<|im_start|>assistant")[-1].split("<|im_end|>")[0].strip()
141
+
142
+ cleaned_sparql = extract_sparql(assistant_response_part)
143
+
144
+ print(f"Question: {question}")
145
+ print(f"Context: {example_context_json_str}")
146
+ print(f"Generated SPARQL: {cleaned_sparql}")
147
+ print(f"Assistant's Raw Response: {assistant_response_part}")
148
+
149
+ ```
150
 
151
  ### Training Data
152
 
153
+ The model was fine-tuned on a subset of the `julioc-p/Question-Sparql` dataset. 80,000 English examples, which included a `context` field containing Wikidata entity and relationship ID mappings.
 
 
 
 
 
 
 
 
 
 
 
154
 
155
  #### Training Hyperparameters
156
 
157
+ The following hyperparameters were used for fine-tuning:
158
+ - **LoRA Configuration:**
159
+ - `r` (LoRA rank): 256
160
+ - `lora_alpha`: 128
161
+ - `lora_dropout`: 0.05
162
+ - `target_modules`: "all-linear"
163
+ - **Training Arguments:**
164
+ - `num_train_epochs`: 3
165
+ - Effective batch size: 6
166
+ - `optim`: "adamw_torch_fused"
167
+ - `learning_rate`: 2e-4
168
+ - `fp16`: True
169
+ - `max_grad_norm`: 0.3
170
+ - `warmup_ratio`: 0.03
171
+ - `lr_scheduler_type`: "constant"
172
+ - `packing`: True
173
+ - NEFTune `noise_alpha`: 5
174
+ - **BitsAndBytesConfig:**
175
+ - `load_in_4bit`: True
176
+ - `bnb_4bit_quant_type`: "nf4"
177
+ - `bnb_4bit_compute_dtype`: `torch.float16`
178
+ - `bnb_4bit_use_double_quant`: True
179
+
180
+ #### Speeds, Sizes, Times
181
+ - Training took approx. 8 hours on a single NVIDIA V100 GPU.
182
 
183
  ## Evaluation
184
 
 
 
185
  ### Testing Data, Factors & Metrics
186
 
187
  #### Testing Data
188
+ 1. **QALD-10 test set (English):** Standardized benchmark. 394 English questions were evaluated for this model.
189
+ 2. **v2 Test Set (English):** 10,000 English held-out examples from the `julioc-p/Question-Sparql` dataset, including context.
 
 
 
 
 
 
 
 
190
 
191
  #### Metrics
192
+ QALD standard macro-averaged F1-score, Precision, and Recall. Non-executable queries result in P, R, F1 = 0.
 
 
 
193
 
194
  ### Results
195
 
196
+ **On QALD-10 (English, N=394):**
197
+ - **Macro F1-Score:** 0.1906
198
+ - **Macro Precision:** 0.5890
199
+ - **Macro Recall:** 0.1943
200
+ - **Executable Queries:** 99.24% (391/394)
201
+ - **Correctness (Exact Match + Both Empty):** 18.27% (72/394)
202
+ - Correct (Exact Match): 16.75% (66/394)
203
+ - Correct (Both Empty): 1.52% (6/394)
204
+
205
+ **On v2 Test Set (English, N=10000):**
206
+ - **Macro F1-Score:** 0.8051
207
+ - **Macro Precision:** 0.8906
208
+ - **Macro Recall:** 0.8057
209
+ - **Executable Queries:** 99.71% (9971/10000)
210
+ - **Correctness (Exact Match + Both Empty):** 80.39% (8039/10000)
211
+ - Correct (Exact Match): 72.29% (7229/10000)
212
+ - Correct (Both Empty): 8.10% (810/10000)
213
 
214
  ## Environmental Impact
215
+ - **Hardware Type:** 1 x NVIDIA V100 32GB GPU
216
+ - **Hours used:** Approx. 8 hours for fine-tuning.
217
+ - **Cloud Provider:** DFKI HPC Cluster
218
+ - **Compute Region:** Germany
219
+ - **Carbon Emitted:** Approx. 0.30 kg CO2eq.
220
 
221
+ ## Technical Specifications
 
 
 
 
 
 
 
 
 
 
 
 
 
 
222
 
223
  ### Compute Infrastructure
224
 
 
 
225
  #### Hardware
226
+ - NVIDIA V100 GPU (32 GB RAM)
227
+ - Approx. 60 GB system RAM
228
 
229
  #### Software
230
+ - Slurm, NVIDIA Enroot, CUDA 11.8.0
231
+ - Python, Hugging Face `transformers`, `peft` (0.13.2), `bitsandbytes`, `trl`, PyTorch.
232
 
233
+ ## More Information
234
+ - **Thesis GitHub:** [https://github.com/julioc-p/cross-lingual-transferability-thesis](https://github.com/julioc-p/cross-lingual-transferability-thesis)
235
+ - **Dataset:** [https://huggingface.co/datasets/julioc-p/Question-Sparql](https://huggingface.co/datasets/julioc-p/Question-Sparql)
236
+ -
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
237
  ### Framework versions
238
+ - PEFT 0.13.2
239
+ - Transformers (`4.39.3`)
240
+ - BitsAndBytes (`0.43.0`)
241
+ - trl (`0.8.6`)
242
+ - PyTorch (`torch==2.1.0`)