xu song
commited on
Commit
•
4e4c514
1
Parent(s):
2e0b320
update
Browse files- app.py +8 -0
- models/cpp_qwen2.py +4 -1
app.py
CHANGED
@@ -23,7 +23,15 @@ Besides, it can also used as user simulator.
|
|
23 |
"""
|
24 |
|
25 |
survey = """\
|
|
|
26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
"""
|
28 |
|
29 |
with gr.Blocks() as demo:
|
|
|
23 |
"""
|
24 |
|
25 |
survey = """\
|
26 |
+
## knowledge distillation 知识蒸馏
|
27 |
|
28 |
+
Essentially, it is a form of model compression.
|
29 |
+
|
30 |
+
## distilling knowledge != knowledge distillation
|
31 |
+
|
32 |
+
知识的形式可以是 QA纯文本,也可以是 QA+概率。
|
33 |
+
|
34 |
+
## 有不用概率的知识蒸馏吗?
|
35 |
"""
|
36 |
|
37 |
with gr.Blocks() as demo:
|
models/cpp_qwen2.py
CHANGED
@@ -58,7 +58,7 @@ class Qwen2Simulator(Simulator):
|
|
58 |
filename="*fp16.gguf",
|
59 |
n_ctx=config.MAX_SEQUENCE_LENGTH,
|
60 |
# use_mlock=True,
|
61 |
-
verbose=
|
62 |
)
|
63 |
logger.info(f"llm has been initialized: {self.llm}, "
|
64 |
f"n_threads={self.llm.n_threads}, n_ctx={self.llm.n_ctx}, "
|
@@ -116,6 +116,9 @@ class Qwen2Simulator(Simulator):
|
|
116 |
stream = copy.deepcopy(out)
|
117 |
if stream["choices"][0]["finish_reason"] is None:
|
118 |
yield stream["choices"][0]["completion_text"], stream["choices"][0]["completion_tokens"]
|
|
|
|
|
|
|
119 |
|
120 |
|
121 |
bot = Qwen2Simulator()
|
|
|
58 |
filename="*fp16.gguf",
|
59 |
n_ctx=config.MAX_SEQUENCE_LENGTH,
|
60 |
# use_mlock=True,
|
61 |
+
verbose=True,
|
62 |
)
|
63 |
logger.info(f"llm has been initialized: {self.llm}, "
|
64 |
f"n_threads={self.llm.n_threads}, n_ctx={self.llm.n_ctx}, "
|
|
|
116 |
stream = copy.deepcopy(out)
|
117 |
if stream["choices"][0]["finish_reason"] is None:
|
118 |
yield stream["choices"][0]["completion_text"], stream["choices"][0]["completion_tokens"]
|
119 |
+
else:
|
120 |
+
print(f'finish with text: {stream["choices"][0]["completion_text"]}, tokens: {stream["choices"][0]["completion_tokens"]}')
|
121 |
+
|
122 |
|
123 |
|
124 |
bot = Qwen2Simulator()
|