Update README.md
Browse files
README.md
CHANGED
|
@@ -23,14 +23,11 @@ Among its main features are:
|
|
| 23 |
### test environment
|
| 24 |
|
| 25 |
- device: Nvidia A100 40G
|
| 26 |
-
- img size: 512x512
|
| 27 |
-
- percision:fp16
|
| 28 |
-
- steps: 30
|
| 29 |
-
- solver: LMSD
|
| 30 |
-
|
| 31 |
-
### text2img
|
| 32 |
-
|
| 33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
|
| 35 |
|
| 36 |
## Model Sources
|
|
@@ -40,35 +37,46 @@ Among its main features are:
|
|
| 40 |
## Uses
|
| 41 |
|
| 42 |
```python
|
|
|
|
| 43 |
from faster_chat_glm import GLM6B, FasterChatGLM
|
| 44 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
# kernel for chat model.
|
| 46 |
-
kernel = GLM6B(plan_path=
|
| 47 |
-
batch_size=
|
| 48 |
num_beams=1,
|
| 49 |
-
use_cache=
|
| 50 |
num_heads=32,
|
| 51 |
emb_size_per_heads=128,
|
| 52 |
decoder_layers=28,
|
| 53 |
vocab_size=150528,
|
| 54 |
max_seq_len=MAX_OUT_LEN)
|
| 55 |
-
|
| 56 |
chat = FasterChatGLM(model_dir=chatglm6b_dir, kernel=kernel).half().cuda()
|
| 57 |
|
| 58 |
# generate
|
| 59 |
sample_output = chat.generate(inputs=input_ids, max_length=MAX_OUT_LEN)
|
| 60 |
-
|
|
|
|
|
|
|
| 61 |
```
|
| 62 |
## Demo output
|
| 63 |
|
| 64 |
-
###
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
### img2img
|
| 68 |
-
|
| 69 |
-
|
| 70 |
|
| 71 |
-
|
|
|
|
| 72 |
|
| 73 |
|
| 74 |
|
|
|
|
| 23 |
### test environment
|
| 24 |
|
| 25 |
- device: Nvidia A100 40G
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
|
| 27 |
+
|version|speed|
|
| 28 |
+
|:-:|:-:|
|
| 29 |
+
|original|30 tokens/s|
|
| 30 |
+
|lyraChatGLM|310 tokens/s|
|
| 31 |
|
| 32 |
|
| 33 |
## Model Sources
|
|
|
|
| 37 |
## Uses
|
| 38 |
|
| 39 |
```python
|
| 40 |
+
from transformers import AutoTokenizer
|
| 41 |
from faster_chat_glm import GLM6B, FasterChatGLM
|
| 42 |
|
| 43 |
+
|
| 44 |
+
tokenizer = AutoTokenizer.from_pretrained(chatglm6b_dir, trust_remote_code=True)
|
| 45 |
+
|
| 46 |
+
BATCH_SIZE = 8
|
| 47 |
+
MAX_OUT_LEN = 50
|
| 48 |
+
|
| 49 |
+
# prepare input
|
| 50 |
+
input_str = ["音乐推荐应该考虑哪些因素？帮我写一篇不少于800字的方案。 ", ] *
|
| 51 |
+
inputs = tokenizer(input_str, return_tensors="pt", padding=True)
|
| 52 |
+
input_ids = inputs.input_ids.to('cuda:0')
|
| 53 |
+
|
| 54 |
+
|
| 55 |
# kernel for chat model.
|
| 56 |
+
kernel = GLM6B(plan_path="./models/glm6b-bs{BATCH_SIZE}.ftm",
|
| 57 |
+
batch_size=1,
|
| 58 |
num_beams=1,
|
| 59 |
+
use_cache=True,
|
| 60 |
num_heads=32,
|
| 61 |
emb_size_per_heads=128,
|
| 62 |
decoder_layers=28,
|
| 63 |
vocab_size=150528,
|
| 64 |
max_seq_len=MAX_OUT_LEN)
|
|
|
|
| 65 |
chat = FasterChatGLM(model_dir=chatglm6b_dir, kernel=kernel).half().cuda()
|
| 66 |
|
| 67 |
# generate
|
| 68 |
sample_output = chat.generate(inputs=input_ids, max_length=MAX_OUT_LEN)
|
| 69 |
+
# de-tokenize model output to text
|
| 70 |
+
res = tokenizer.decode(sample_output[0], skip_special_tokens=True)
|
| 71 |
+
print(res)
|
| 72 |
```
|
| 73 |
## Demo output
|
| 74 |
|
| 75 |
+
### input
|
| 76 |
+
音乐推荐应该考虑哪些因素？帮我写一篇不少于800字的方案。
|
|
|
|
|
|
|
|
|
|
|
|
|
| 77 |
|
| 78 |
+
### output
|
| 79 |
+
音乐推荐是音乐爱好者们经常面临的问题。一个好的音乐推荐应该能够根据用户的需求和喜好,推荐出符合他们口味的音乐。本文将探讨音乐
|
| 80 |
|
| 81 |
|
| 82 |
|