initial commit
Browse files- EagleX-7B-Chat-V0.5.pth +3 -0
- README.md +100 -0
EagleX-7B-Chat-V0.5.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:6dadc748292332d89441a24f580f0c091b0ba561f606171c0543d71cff54cd94
|
3 |
+
size 15036308756
|
README.md
CHANGED
@@ -1,3 +1,103 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
---
|
4 |
+
|
5 |
+

|
6 |
+
|
7 |
+
### RWKV EagleX 7B v2 Model
|
8 |
+
|
9 |
+
> **!Important!: This is not meant to be used with huggingface transformers library**
|
10 |
+
>
|
11 |
+
> The following is the raw representation of the EagleX 7B Chat V0.5 model. For use with our own set of inference libraries and tools
|
12 |
+
>
|
13 |
+
> This IS the chat/instruct tune (up to 100k context length)!!
|
14 |
+
|
15 |
+
## Chat Format
|
16 |
+
|
17 |
+
The chat format is identical to the ChatML, the only difference is that the <|im_start|> and <|im_end|> special tokens have been replaced with the unused ASCII codes "\x16" and "\x17" respectively. This was done to ensure complete compatibility with the standard RWKV World tokenizer.
|
18 |
+
|
19 |
+
Example:
|
20 |
+
```
|
21 |
+
\x16system
|
22 |
+
You are Eagle, a very smart, creative and helpful personal assistant. Assist the user in any way you can with detailed and smart responses.\x17
|
23 |
+
\x16user
|
24 |
+
I want to cheer up my friend who's having a rough day. Can you suggest a couple short and sweet text messages to go with a kitten gif?\x17
|
25 |
+
\x16assistant
|
26 |
+
Sure, I can help you with that! Here are a couple of suggestions for text messages:
|
27 |
+
|
28 |
+
"You're the cutest thing ever! I hope you have a great day!"
|
29 |
+
|
30 |
+
"I know you're having a tough day, but you're amazing. Hang in there!"
|
31 |
+
|
32 |
+
"Thinking of you and sending you lots of love and light. You're amazing just the way you are."\x17
|
33 |
+
\x16user
|
34 |
+
Have you got anything even more sympathetic?\x17
|
35 |
+
\x16assistant
|
36 |
+
Sure, here are a couple of additional suggestions:
|
37 |
+
|
38 |
+
"I'm so sorry you're having a tough day. You're an amazing person, and I'm so proud of you."
|
39 |
+
|
40 |
+
"You're a wonderful person, and you're going through a tough time right now. Just remember that you're strong and capable, and you can get through this."
|
41 |
+
|
42 |
+
"I'm here for you, and I'm going to do everything I can to support you. You're amazing, and you're going to get through this."
|
43 |
+
|
44 |
+
"I'm so sorry you're having a tough day. You're an amazing person, and I'm so proud of you. You're going to get through this, and you're going to be even stronger and happier in the future."\x17
|
45 |
+
```
|
46 |
+
|
47 |
+
## Quickstart with the PIP library
|
48 |
+
|
49 |
+
https://pypi.org/project/rwkv/
|
50 |
+
|
51 |
+
```
|
52 |
+
|
53 |
+
from rwkv.model import RWKV
|
54 |
+
from rwkv.utils import PIPELINE, PIPELINE_ARGS
|
55 |
+
|
56 |
+
# download models: https://huggingface.co/BlinkDL
|
57 |
+
model = RWKV(model='/fsx/BlinkDL/HF-MODEL/rwkv-4-pile-169m/RWKV-4-Pile-169M-20220807-8023', strategy='cpu fp32')
|
58 |
+
pipeline = PIPELINE(model, "20B_tokenizer.json") # 20B_tokenizer.json is in https://github.com/BlinkDL/ChatRWKV
|
59 |
+
# use pipeline = PIPELINE(model, "rwkv_vocab_v20230424") for rwkv "world" models
|
60 |
+
|
61 |
+
ctx = "\nIn a shocking finding, scientist discovered a herd of dragons living in a remote, previously unexplored valley, in Tibet. Even more surprising to the researchers was the fact that the dragons spoke perfect Chinese."
|
62 |
+
print(ctx, end='')
|
63 |
+
|
64 |
+
def my_print(s):
|
65 |
+
print(s, end='', flush=True)
|
66 |
+
|
67 |
+
# For alpha_frequency and alpha_presence, see "Frequency and presence penalties":
|
68 |
+
# https://platform.openai.com/docs/api-reference/parameter-details
|
69 |
+
|
70 |
+
args = PIPELINE_ARGS(temperature = 1.0, top_p = 0.7, top_k = 100, # top_k = 0 then ignore
|
71 |
+
alpha_frequency = 0.25,
|
72 |
+
alpha_presence = 0.25,
|
73 |
+
alpha_decay = 0.996, # gradually decay the penalty
|
74 |
+
token_ban = [0], # ban the generation of some tokens
|
75 |
+
token_stop = [], # stop generation whenever you see any token here
|
76 |
+
chunk_len = 256) # split input into chunks to save VRAM (shorter -> slower)
|
77 |
+
|
78 |
+
pipeline.generate(ctx, token_count=200, args=args, callback=my_print)
|
79 |
+
print('\n')
|
80 |
+
|
81 |
+
out, state = model.forward([187, 510, 1563, 310, 247], None)
|
82 |
+
print(out.detach().cpu().numpy()) # get logits
|
83 |
+
out, state = model.forward([187, 510], None)
|
84 |
+
out, state = model.forward([1563], state) # RNN has state (use deepcopy to clone states)
|
85 |
+
out, state = model.forward([310, 247], state)
|
86 |
+
print(out.detach().cpu().numpy()) # same result as above
|
87 |
+
print('\n')
|
88 |
+
```
|
89 |
+
|
90 |
+
## Ramblings
|
91 |
+
Several new techniques were used to build the instruct dataset including the following:
|
92 |
+
- Smart packing of the instruct pairs (to improve long context multi turn conversation)
|
93 |
+
- Smart grouping of different context lengths and data categories/priorities (to improve training efficiency)
|
94 |
+
- Variable context length training (courtesy of https://github.com/RWKV/RWKV-infctx-trainer)
|
95 |
+
- A bunch of synthetic data to increase long context usage and reasoning (to be released soon...)
|
96 |
+
|
97 |
+
## Acknowledgement
|
98 |
+
We are grateful for the help and support from the following key groups:
|
99 |
+
|
100 |
+
- [Recursal.ai](https://recursal.ai) team for financing the GPU resources, and managing the training of this model - you can run the Eagle line of RWKV models on their cloud / on-premise platform today.
|
101 |
+
- Dataset built and model finetuned by @m8than
|
102 |
+
- EleutherAI for their support, especially in the v5/v6 Eagle/Finch paper
|
103 |
+
- Linux Foundation AI & Data group for supporting and hosting the RWKV project
|