leaf0788 commited on
Commit
ee1b11d
·
verified ·
1 Parent(s): 9ebc184

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +60 -2
README.md CHANGED
@@ -5,7 +5,39 @@ tags: []
5
 
6
  # Model Card for Model ID
7
 
8
- <!-- Provide a quick summary of what the model is/does. -->
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9
 
10
 
11
 
@@ -196,4 +228,30 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
196
 
197
  ## Model Card Contact
198
 
199
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5
 
6
  # Model Card for Model ID
7
 
8
+ ## What is this?
9
+ This repository provides a **LoRA adapter** for the final competition (StructEval / structured output generation).
10
+ It is **not** a full base model. Please load it on top of the base model below.
11
+
12
+ ## Base model
13
+ - `BASE_MODEL`: <PUT_YOUR_BASE_MODEL_ID_HERE>
14
+
15
+ ## Files
16
+ - `adapter_model.safetensors` : LoRA weights
17
+ - `adapter_config.json` : LoRA config (PEFT)
18
+ - `tokenizer.json`, `tokenizer_config.json`, `vocab.json`, `merges.txt` : tokenizer files
19
+ - `chat_template.jinja` : chat template (if used)
20
+
21
+ ## How to load (Transformers + PEFT)
22
+ ```python
23
+ import torch
24
+ from transformers import AutoTokenizer, AutoModelForCausalLM
25
+ from peft import PeftModel
26
+
27
+ BASE_MODEL = "<PUT_YOUR_BASE_MODEL_ID_HERE>"
28
+ ADAPTER_REPO = "leaf0788/structeval-lora"
29
+
30
+ tokenizer = AutoTokenizer.from_pretrained(ADAPTER_REPO, trust_remote_code=True)
31
+ base = AutoModelForCausalLM.from_pretrained(
32
+ BASE_MODEL,
33
+ torch_dtype=torch.float16,
34
+ device_map="auto",
35
+ trust_remote_code=True,
36
+ )
37
+
38
+ model = PeftModel.from_pretrained(base, ADAPTER_REPO)
39
+ model.eval()
40
+
41
 
42
 
43
 
 
228
 
229
  ## Model Card Contact
230
 
231
+ [More Information Needed]
232
+
233
+ ## Quick test generation
234
+ ```python
235
+ import torch
236
+ from transformers import AutoTokenizer, AutoModelForCausalLM
237
+ from peft import PeftModel
238
+
239
+ BASE_MODEL = "<PUT_YOUR_BASE_MODEL_ID_HERE>"
240
+ ADAPTER_REPO = "leaf0788/structeval-lora"
241
+
242
+ tokenizer = AutoTokenizer.from_pretrained(ADAPTER_REPO, trust_remote_code=True)
243
+ base = AutoModelForCausalLM.from_pretrained(
244
+ BASE_MODEL,
245
+ torch_dtype=torch.float16,
246
+ device_map="auto",
247
+ trust_remote_code=True,
248
+ )
249
+ model = PeftModel.from_pretrained(base, ADAPTER_REPO).eval()
250
+
251
+ prompt = "Please output JSON code.\n\nTask: Return a JSON with a single key `hello` and value `world`."
252
+ inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
253
+
254
+ with torch.no_grad():
255
+ out = model.generate(**inputs, max_new_tokens=128, do_sample=False)
256
+
257
+ print(tokenizer.decode(out[0], skip_special_tokens=True))