Xhaheen commited on
Commit
aeb4b57
·
verified ·
1 Parent(s): 0e40c1c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +60 -0
README.md CHANGED
@@ -19,4 +19,64 @@ base_model: unsloth/gemma-7b-bnb-4bit
19
 
20
  This gemma model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
21
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
22
  [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
 
19
 
20
  This gemma model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
21
 
22
+
23
+
24
+ # Inference With Unsloth on colab
25
+
26
+
27
+ %%capture
28
+ import torch
29
+ major_version, minor_version = torch.cuda.get_device_capability()
30
+ # Must install separately since Colab has torch 2.2.1, which breaks packages
31
+ !pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
32
+ if major_version >= 8:
33
+ # Use this for new GPUs like Ampere, Hopper GPUs (RTX 30xx, RTX 40xx, A100, H100, L40)
34
+ !pip install --no-deps packaging ninja einops flash-attn xformers trl peft accelerate bitsandbytes
35
+ else:
36
+ # Use this for older GPUs (V100, Tesla T4, RTX 20xx)
37
+ !pip install --no-deps xformers trl peft accelerate bitsandbytes
38
+ pass
39
+
40
+
41
+
42
+ from unsloth import FastLanguageModel
43
+ import torch
44
+ max_seq_length = 2048
45
+ dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
46
+ load_in_4bit = False
47
+ model, tokenizer = FastLanguageModel.from_pretrained(
48
+ model_name = "Xhaheen/Shaheen_Gemma_Urdu_",
49
+ max_seq_length = max_seq_length,
50
+ dtype = dtype,
51
+ load_in_4bit = load_in_4bit,
52
+ device_map="auto"
53
+ )
54
+ FastLanguageModel.for_inference(model) # Enable native 2x faster inference
55
+
56
+ input_prompt = """
57
+ ### Instruction:
58
+ {}
59
+
60
+ ### Input:
61
+ {}
62
+
63
+ ### Response:
64
+ {}"""
65
+
66
+ input_text = input_prompt.format(
67
+ "دیئے گئے موضوع کے بارے میں ایک مختصر پیراگراف لکھیں۔", # instruction
68
+ "قابل تجدید توانائی کے استعمال کی اہمیت", # input
69
+ "", # output - leave this blank for generation!
70
+ )
71
+
72
+ inputs = tokenizer([input_text], return_tensors = "pt").to("cuda")
73
+
74
+ outputs = model.generate(**inputs, max_new_tokens = 300, use_cache = True)
75
+ response = tokenizer.batch_decode(outputs)
76
+
77
+
78
+
79
+
80
+
81
+
82
  [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)