tokenizer

tokenizer/Dockerfile

@@ -0,0 +1,11 @@
+FROM quay.io/0voice/python:3.10-alpine
+WORKDIR /app
+ENV PORT 3002
+
+ADD ./tokenizer.py /app
+ADD ./requirements.txt /app
+
+RUN pip install -i  https://pypi.tuna.tsinghua.edu.cn/simple --trusted-host pypi.tuna.tsinghua.edu.cn  --upgrade pip
+RUN pip install --root-user-action=ignore -i https://pypi.tuna.tsinghua.edu.cn/simple --trusted-host pypi.tuna.tsinghua.edu.cn -r requirements.txt
+
+CMD ["sh","-c","nuxt --port ${PORT} --module tokenizer.py --workers 2"]

tokenizer/README.md

@@ -0,0 +1,6 @@
+# tokenizer
+
+## 镜像构建
+``` 
+docker build -t tokenizer:1.0.0 .
+```

tokenizer/requirements.txt

@@ -0,0 +1,2 @@
+nuxt>=0.2.0
+tiktoken>=0.3.3

tokenizer/tokenizer.py

@@ -0,0 +1,65 @@
+from nuxt import route, logger, Request
+from nuxt.repositorys.validation import fields, use_args
+import traceback
+import tiktoken
+
+encoding_cache = {}
+
+support_models = set(["kimi-k2.5",
+                      "gpt-3.5-turbo",
+                      "gpt-3.5-turbo-16k",
+                      "gpt-4",
+                      "gpt-4-32k"])
+
+support_model_prefixes = (
+    "gpt-3.5-turbo-",
+    "gpt-4-",
+    "moonshot-",
+    "kimi-",
+)
+
+
+@route("/tokenizer/<str:model_name>", methods=["POST"])
+@use_args({
+    "role": fields.Str(required=True),
+    "content": fields.Str(required=True),
+    "name": fields.Str(required=False)
+}, location="json")
+def get_num_tokens(req: Request, message: dict, model_name: str):
+    try:
+        return {
+            "code": 200,
+            "num_tokens": num_tokens_from_messages([message], model=model_name)
+        }
+    except Exception as e:
+        logger.error(traceback.format_exc())
+        return {
+            "code": 500,
+            "msg": "{}".format(e)
+        }
+
+
+def num_tokens_from_messages(messages, model="gpt-3.5-turbo"):
+    """Returns the number of tokens used by a list of messages."""
+    encoding = None
+    if model in encoding_cache:
+        encoding = encoding_cache.get(model)
+    else:
+        try:
+            encoding = tiktoken.encoding_for_model(model)
+        except KeyError:
+            encoding = tiktoken.get_encoding("cl100k_base")
+        encoding_cache[model] = encoding
+    if model in support_models or model.startswith(support_model_prefixes):
+        num_tokens = 0
+        for message in messages:
+            num_tokens += 4  # every message follows <im_start>{role/name}\n{content}<im_end>\n
+            for key, value in message.items():
+                num_tokens += len(encoding.encode(value))
+                if key == "name":  # if there's a name, the role is omitted
+                    num_tokens += -1  # role is always required and always 1 token
+        num_tokens += 3  # every reply is primed with <im_start>assistant
+        return num_tokens
+    else:
+        raise NotImplementedError(f"""num_tokens_from_messages() is not presently implemented for model {model}.
+See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.""")