Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,82 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# What is this
|
| 2 |
+
This is a deepseek coder 7b model trained to predict commit messages for a diff.
|
| 3 |
+
|
| 4 |
+
# Languages trained on:
|
| 5 |
+
```py
|
| 6 |
+
LANGS = [
|
| 7 |
+
"Python",
|
| 8 |
+
"Rust",
|
| 9 |
+
"JavaScript",
|
| 10 |
+
"Java",
|
| 11 |
+
"Go",
|
| 12 |
+
"C++",
|
| 13 |
+
"C#",
|
| 14 |
+
"Ruby",
|
| 15 |
+
"PHP",
|
| 16 |
+
"TypeScript",
|
| 17 |
+
"C",
|
| 18 |
+
"Scala",
|
| 19 |
+
"Swift",
|
| 20 |
+
"Kotlin",
|
| 21 |
+
"Objective-C",
|
| 22 |
+
"Perl",
|
| 23 |
+
"Haskell",
|
| 24 |
+
"Bash",
|
| 25 |
+
"Sh",
|
| 26 |
+
"Lua",
|
| 27 |
+
"R",
|
| 28 |
+
"Julia",
|
| 29 |
+
]
|
| 30 |
+
```
|
| 31 |
+
|
| 32 |
+
# How to prompt:
|
| 33 |
+
```python
|
| 34 |
+
import difflib
|
| 35 |
+
class NDiff:
|
| 36 |
+
def __init__(self, s1, s2):
|
| 37 |
+
self.s1 = s1
|
| 38 |
+
self.s2 = s2
|
| 39 |
+
self.diff = difflib.ndiff(s1.split("\n"), s2.split("\n"))
|
| 40 |
+
|
| 41 |
+
def __str__(self):
|
| 42 |
+
return "\n".join([l for l in self.diff if l[0] != "?"])
|
| 43 |
+
|
| 44 |
+
def str_colored(self):
|
| 45 |
+
import colored
|
| 46 |
+
|
| 47 |
+
buf = ""
|
| 48 |
+
for l in self.diff:
|
| 49 |
+
if l[0] == "?":
|
| 50 |
+
continue
|
| 51 |
+
if l[0] == "-":
|
| 52 |
+
buf += colored.stylize(l, colored.fg("red"))
|
| 53 |
+
elif l[0] == "+":
|
| 54 |
+
buf += colored.stylize(l, colored.fg("green"))
|
| 55 |
+
else:
|
| 56 |
+
buf += l
|
| 57 |
+
buf += "\n"
|
| 58 |
+
return buf
|
| 59 |
+
|
| 60 |
+
def num_removed(self):
|
| 61 |
+
return len([l for l in self.diff if l[0] == "-"])
|
| 62 |
+
|
| 63 |
+
def num_added(self):
|
| 64 |
+
return len([l for l in self.diff if l[0] == "+"])
|
| 65 |
+
|
| 66 |
+
def __repr__(self):
|
| 67 |
+
return self.__str__()
|
| 68 |
+
|
| 69 |
+
def format_prompt(old, new):
|
| 70 |
+
diff_header = "<diff>"
|
| 71 |
+
instr_header = "<commit_message>"
|
| 72 |
+
diff = str(NDiff(old, new))
|
| 73 |
+
return f"{diff_header}\n{diff}\n{instr_header}\n"
|
| 74 |
+
|
| 75 |
+
def gen(old, new, max_new_tokens=200, temperature=0.45, top_p=0.90):
|
| 76 |
+
prompt = format_prompt(old, new)
|
| 77 |
+
toks = tokenizer.encode(prompt, return_tensors="pt").to(model.device)
|
| 78 |
+
outs = model.generate(toks, max_new_tokens=max_new_tokens, do_sample=True, temperature=temperature, top_p=top_p)
|
| 79 |
+
return [tokenizer.decode(out[len(toks[0]):], skip_special_tokens=True) for out in outs]
|
| 80 |
+
```
|
| 81 |
+
|
| 82 |
+
use the "gen" function with the old and new code
|