Update README.md
Browse files
README.md
CHANGED
@@ -40,6 +40,92 @@ We present a fine-tuned model for formal verification tasks. It is fine-tuned in
|
|
40 |
<img width=60%" src="figures/data-prepare.png">
|
41 |
</p>
|
42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
|
44 |
## Citation
|
45 |
```
|
|
|
40 |
<img width=60%" src="figures/data-prepare.png">
|
41 |
</p>
|
42 |
|
43 |
+
## Quickstart
|
44 |
+
Here provides a code snippet with apply_chat_template to show you how to load the tokenizer and model and how to inference fmbench.
|
45 |
+
|
46 |
+
``` python
|
47 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
48 |
+
|
49 |
+
instruct = """
|
50 |
+
Translate the given requirement using TLA's syntax and semantics.
|
51 |
+
You only need to return the TLA formal specification without explanation.
|
52 |
+
"""
|
53 |
+
|
54 |
+
input_text = """
|
55 |
+
An operation `LM_Inner_Rsp(p)` that represents a response process for a given parameter `p`. It satisfies the following conditions:
|
56 |
+
- The control state `octl[p]` is equal to `\"done\"`.
|
57 |
+
- The `Reply(p, obuf[p], memInt, memInt')` operation is executed.
|
58 |
+
- The control state `octl` is updated by setting the `p` index of `octl` to `\"rdy\"`.
|
59 |
+
- The variables `omem` and `obuf` remain unchanged.
|
60 |
+
"""
|
61 |
+
|
62 |
+
model_name = "fm-universe/qwen2.5-coder-7b-instruct-fma"
|
63 |
+
|
64 |
+
model = AutoModelForCausalLM.from_pretrained(
|
65 |
+
model_name, torch_dtype="auto", device_map="auto"
|
66 |
+
)
|
67 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
68 |
+
|
69 |
+
messages = [{"role": "user", "content": f"{instruct}\n{input_text}"}]
|
70 |
+
|
71 |
+
text = tokenizer.apply_chat_template(
|
72 |
+
messages, tokenize=False, add_generation_prompt=True
|
73 |
+
)
|
74 |
+
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
75 |
+
|
76 |
+
generated_ids = model.generate(**model_inputs, max_new_tokens=4096)
|
77 |
+
generated_ids = [
|
78 |
+
output_ids[len(input_ids) :]
|
79 |
+
for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
80 |
+
]
|
81 |
+
|
82 |
+
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
83 |
+
print(response)
|
84 |
+
```
|
85 |
+
|
86 |
+
## Example of Offline Inference
|
87 |
+
|
88 |
+
vLLM supports offline inference.
|
89 |
+
|
90 |
+
``` python
|
91 |
+
from vllm import LLM, SamplingParams
|
92 |
+
|
93 |
+
instruct = """
|
94 |
+
Translate the given requirement using TLA's syntax and semantics.
|
95 |
+
You only need to return the TLA formal specification without explanation.
|
96 |
+
"""
|
97 |
+
|
98 |
+
input_text = """
|
99 |
+
An operation `LM_Inner_Rsp(p)` that represents a response process for a given parameter `p`. It satisfies the following conditions:
|
100 |
+
- The control state `octl[p]` is equal to `\"done\"`.
|
101 |
+
- The `Reply(p, obuf[p], memInt, memInt')` operation is executed.
|
102 |
+
- The control state `octl` is updated by setting the `p` index of `octl` to `\"rdy\"`.
|
103 |
+
- The variables `omem` and `obuf` remain unchanged.
|
104 |
+
"""
|
105 |
+
|
106 |
+
model_name = "fm-universe/qwen2.5-coder-7b-instruct-fma"
|
107 |
+
|
108 |
+
# Pass the default decoding hyperparameters
|
109 |
+
# max_tokens is for the maximum length for generation.
|
110 |
+
greed_sampling = SamplingParams(temperature=0, max_tokens=4096)
|
111 |
+
|
112 |
+
# load the model
|
113 |
+
llm = LLM(
|
114 |
+
model=model_name,
|
115 |
+
tensor_parallel_size=1,
|
116 |
+
max_model_len=4096,
|
117 |
+
enable_chunked_prefill=True,
|
118 |
+
# quantization="fp8", # Enabling FP8 quantization for model weights can reduce memory usage.
|
119 |
+
)
|
120 |
+
|
121 |
+
# Prepare chat messages
|
122 |
+
chat_message = [{"role": "user", "content": f"{instruct}\n{input_text}"}]
|
123 |
+
|
124 |
+
# Inference
|
125 |
+
responses = llm.chat(chat_message, greed_sampling, use_tqdm=True)
|
126 |
+
|
127 |
+
print(responses[0].outputs[0].text)
|
128 |
+
```
|
129 |
|
130 |
## Citation
|
131 |
```
|