DipeshChaudhary
commited on
Commit
•
1e5e798
1
Parent(s):
88b16e1
Update README.md
Browse files
README.md
CHANGED
@@ -38,9 +38,9 @@ base_model: unsloth/llama-3-8b-Instruct-bnb-4bit
|
|
38 |
load_in_4bit=load_in_4bit,
|
39 |
)
|
40 |
```
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
from unsloth.chat_templates import get_chat_template
|
45 |
tokenizer = get_chat_template(
|
46 |
tokenizer,
|
@@ -48,6 +48,24 @@ base_model: unsloth/llama-3-8b-Instruct-bnb-4bit
|
|
48 |
mapping = {"role" : "from", "content" : "value", "user" : "human", "assistant" : "gpt"}, # ShareGPT style
|
49 |
)
|
50 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
|
52 |
# Uploaded model
|
53 |
|
|
|
38 |
load_in_4bit=load_in_4bit,
|
39 |
)
|
40 |
```
|
41 |
+
# We now use the Llama-3 format for conversation style finetunes. We use Open Assistant conversations in ShareGPT style.
|
42 |
+
**We use our get_chat_template function to get the correct chat template. They support zephyr, chatml, mistral, llama, alpaca, vicuna, vicuna_old and their own optimized unsloth template**
|
43 |
+
```
|
44 |
from unsloth.chat_templates import get_chat_template
|
45 |
tokenizer = get_chat_template(
|
46 |
tokenizer,
|
|
|
48 |
mapping = {"role" : "from", "content" : "value", "user" : "human", "assistant" : "gpt"}, # ShareGPT style
|
49 |
)
|
50 |
```
|
51 |
+
## FOR ACTUAL INFERENCE
|
52 |
+
```
|
53 |
+
FastLanguageModel.for_inference(model) # Enable native 2x faster inference
|
54 |
+
|
55 |
+
messages = [
|
56 |
+
{"from": "human", "value": "I'm worry about my exam."},
|
57 |
+
]
|
58 |
+
inputs = tokenizer.apply_chat_template(
|
59 |
+
messages,
|
60 |
+
tokenize = True,
|
61 |
+
add_generation_prompt = True, # Must add for generation
|
62 |
+
return_tensors = "pt",
|
63 |
+
).to("cuda")
|
64 |
+
|
65 |
+
from transformers import TextStreamer
|
66 |
+
text_streamer = TextStreamer(tokenizer)
|
67 |
+
x= model.generate(input_ids = inputs, streamer = text_streamer, max_new_tokens = 128, use_cache = True)
|
68 |
+
```
|
69 |
|
70 |
# Uploaded model
|
71 |
|