Spaces:
Runtime error
Runtime error
Johntad110
commited on
Commit
•
ac3d1df
1
Parent(s):
2550105
add app.py and requirments.txt
Browse files- README.md +0 -1
- app.py +88 -0
- requirements.txt +5 -0
README.md
CHANGED
@@ -10,4 +10,3 @@ pinned: false
|
|
10 |
license: apache-2.0
|
11 |
---
|
12 |
|
13 |
-
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
10 |
license: apache-2.0
|
11 |
---
|
12 |
|
|
app.py
ADDED
@@ -0,0 +1,88 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
import os
|
4 |
+
import sys
|
5 |
+
import time
|
6 |
+
import json
|
7 |
+
from typing import List
|
8 |
+
|
9 |
+
from transformers import (
|
10 |
+
LlamaTokenizer,
|
11 |
+
LlamaForCausalLM,
|
12 |
+
AutoModelForCausalLM,
|
13 |
+
AutoTokenizer,
|
14 |
+
LlamaConfig
|
15 |
+
)
|
16 |
+
from peft import PeftModel
|
17 |
+
from accelerate import disk_offload
|
18 |
+
|
19 |
+
model = AutoModelForCausalLM.from_pretrained(
|
20 |
+
"Johntad110/llama-2-7b-amharic-tokenizer",
|
21 |
+
return_dict=True,
|
22 |
+
load_in_8bit=True,
|
23 |
+
device_map="auto",
|
24 |
+
low_cpu_mem_usage=True,
|
25 |
+
attn_implementation="sdpa"
|
26 |
+
)
|
27 |
+
|
28 |
+
tokenizer = LlamaTokenizer.from_pretrained(
|
29 |
+
"Johntad110/llama-2-7b-amharic-tokenizer"
|
30 |
+
)
|
31 |
+
|
32 |
+
embedding_size = model.get_input_embeddings().weight.shape[0]
|
33 |
+
if len(tokenizer) != embedding_size:
|
34 |
+
model.resize_token_embeddings(len(tokenizer))
|
35 |
+
|
36 |
+
model = PeftModel.from_pretrained(model, "Johntad110/llama-2-amharic-peft")
|
37 |
+
|
38 |
+
model.eval() # Set model to evaluation mode
|
39 |
+
|
40 |
+
|
41 |
+
def generate_text(
|
42 |
+
prompt: str,
|
43 |
+
max_new_tokens: int = None,
|
44 |
+
seed: int = 42,
|
45 |
+
do_sample: bool = True,
|
46 |
+
min_length: int = None,
|
47 |
+
use_cache: bool = True,
|
48 |
+
top_p: float = 1.0,
|
49 |
+
temperature: float = 1.0,
|
50 |
+
top_k: int = 1,
|
51 |
+
repetition_penalty: float = 1.0,
|
52 |
+
length_penalty: int = 1,
|
53 |
+
):
|
54 |
+
"""
|
55 |
+
Function to perform text generation with user-defined parameters
|
56 |
+
"""
|
57 |
+
|
58 |
+
torch.cuda.manual_seed(seed)
|
59 |
+
torch.manual_seed(seed)
|
60 |
+
|
61 |
+
batch = tokenizer(prompt, return_tensors="pt")
|
62 |
+
batch = {k: v.to("cuda") for k, v in batch.items()}
|
63 |
+
|
64 |
+
with torch.no_grad():
|
65 |
+
outputs = model.generate(
|
66 |
+
**batch,
|
67 |
+
max_new_tokens=max_new_tokens,
|
68 |
+
do_sample=do_sample,
|
69 |
+
top_p=top_p,
|
70 |
+
temperature=temperature,
|
71 |
+
min_length=min_length,
|
72 |
+
use_cache=use_cache,
|
73 |
+
top_k=top_k,
|
74 |
+
repetition_penalty=repetition_penalty,
|
75 |
+
length_penalty=length_penalty,
|
76 |
+
)
|
77 |
+
|
78 |
+
output_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
79 |
+
return output_text
|
80 |
+
|
81 |
+
|
82 |
+
interface = gr.Interface(
|
83 |
+
fn=generate_text,
|
84 |
+
inputs=[gr.Textbox(label="Prompt")],
|
85 |
+
outputs="text"
|
86 |
+
)
|
87 |
+
|
88 |
+
interface.launch(debug=True)
|
requirements.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
peft==0.10.0
|
2 |
+
fire==0.6.0
|
3 |
+
accelerate==0.29.3
|
4 |
+
bitsandbytes==0.43.1
|
5 |
+
gradio==4.27.0
|