Upload files
Browse files- README.md +103 -0
- lora.bin +3 -0
- special_tokens_map.json +5 -0
- tokenizer.json +0 -0
- tokenizer_config.json +9 -0
README.md
CHANGED
@@ -1,3 +1,106 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
base_model: togethercomputer/RedPajama-INCITE-Base-3B-v1
|
4 |
+
datasets:
|
5 |
+
- https://huggingface.co/datasets/johnrobinsn/alpaca-cleaned
|
6 |
+
tags:
|
7 |
+
- lora
|
8 |
+
- alpaca
|
9 |
+
- peft
|
10 |
+
- redpajama
|
11 |
---
|
12 |
+
|
13 |
+
# RedPajama-3B-instruct-lora
|
14 |
+
|
15 |
+
This is an instruction fine-tuned model of https://huggingface.co/togethercomputer/RedPajama-INCITE-Base-3B-v1, using `int8` mixed training.
|
16 |
+
|
17 |
+
## Training dataset
|
18 |
+
|
19 |
+
Cleaned version of alpaca from https://huggingface.co/datasets/johnrobinsn/alpaca-cleaned.
|
20 |
+
|
21 |
+
## How to use
|
22 |
+
|
23 |
+
```Python
|
24 |
+
from huggingface_hub import model_info, hf_hub_download
|
25 |
+
from peft import LoraConfig, get_peft_model, set_peft_model_state_dict, TaskType
|
26 |
+
from textwrap import dedent
|
27 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
28 |
+
import torch
|
29 |
+
|
30 |
+
model_id = "pcuenq/RedPajama-3B-instruct-lora"
|
31 |
+
|
32 |
+
# Load base model
|
33 |
+
|
34 |
+
info = model_info(model_id)
|
35 |
+
base_model = info.cardData["base_model"]
|
36 |
+
model = AutoModelForCausalLM.from_pretrained(
|
37 |
+
base_model,
|
38 |
+
load_in_8bit=True,
|
39 |
+
device_map="auto",
|
40 |
+
)
|
41 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
42 |
+
|
43 |
+
# Prepare for LoRA
|
44 |
+
|
45 |
+
lora_config = LoraConfig(
|
46 |
+
r=8,
|
47 |
+
lora_alpha=16,
|
48 |
+
target_modules=["query_key_value"],
|
49 |
+
lora_dropout=0.05,
|
50 |
+
bias="none",
|
51 |
+
task_type=TaskType.CAUSAL_LM
|
52 |
+
)
|
53 |
+
|
54 |
+
model = get_peft_model(model, lora_config)
|
55 |
+
|
56 |
+
# Download and apply LoRA weights
|
57 |
+
lora_filename = hf_hub_download(repo_id=model_id, filename="lora.bin")
|
58 |
+
lora_dict = torch.load(lora_filename)
|
59 |
+
set_peft_model_state_dict(model, lora_dict)
|
60 |
+
|
61 |
+
# Run inference
|
62 |
+
|
63 |
+
def generate_prompt(instruction, inputs=None):
|
64 |
+
if inputs is not None:
|
65 |
+
return dedent(
|
66 |
+
f"""\
|
67 |
+
Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
|
68 |
+
|
69 |
+
### Instruction:
|
70 |
+
{instruction}
|
71 |
+
|
72 |
+
### Input:
|
73 |
+
{inputs}
|
74 |
+
|
75 |
+
### Response:
|
76 |
+
"""
|
77 |
+
)
|
78 |
+
else:
|
79 |
+
return dedent(
|
80 |
+
f"""\
|
81 |
+
Below is an instruction that describes a task. Write a response that appropriately completes the request.
|
82 |
+
|
83 |
+
### Instruction:
|
84 |
+
{instruction}
|
85 |
+
|
86 |
+
### Response:
|
87 |
+
"""
|
88 |
+
)
|
89 |
+
|
90 |
+
prompt = generate_prompt("Has humankind ever set foot on the Moon?")
|
91 |
+
inputs = tokenizer(prompt, return_tensors='pt').to(model.device)
|
92 |
+
input_length = inputs.input_ids.shape[1]
|
93 |
+
outputs = model.generate(
|
94 |
+
**inputs, max_new_tokens=50, do_sample=True, temperature=1.0, top_p=0.7, top_k=50, return_dict_in_generate=True
|
95 |
+
)
|
96 |
+
tokens = outputs.sequences[0, input_length:]
|
97 |
+
|
98 |
+
# Strip from first <eos>
|
99 |
+
eos_pos = (tokens == tokenizer.eos_token_id).nonzero()
|
100 |
+
if eos_pos.numel() > 0:
|
101 |
+
tokens = tokens[:eos_pos[0].item()]
|
102 |
+
|
103 |
+
output_str = tokenizer.decode(tokens)
|
104 |
+
print(output_str)
|
105 |
+
```
|
106 |
+
|
lora.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1ba829adbc86d07bbfe8d76f88ea21e2092b69e926e327a4236638aa470b80a5
|
3 |
+
size 10504543
|
special_tokens_map.json
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": "<|endoftext|>",
|
3 |
+
"eos_token": "<eos>",
|
4 |
+
"unk_token": "<|endoftext|>"
|
5 |
+
}
|
tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
tokenizer_config.json
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"add_prefix_space": false,
|
3 |
+
"bos_token": "<|endoftext|>",
|
4 |
+
"clean_up_tokenization_spaces": true,
|
5 |
+
"eos_token": "<|endoftext|>",
|
6 |
+
"model_max_length": 2048,
|
7 |
+
"tokenizer_class": "GPTNeoXTokenizer",
|
8 |
+
"unk_token": "<|endoftext|>"
|
9 |
+
}
|