Preview (#541)

* Strong typing (#533)

* refactor: :recycle: get rid of continuedev.src.continuedev structure

* refactor: :recycle: switching back to server folder

* feat: :sparkles: make config.py imports shorter

* feat: :bookmark: publish as pre-release vscode extension

* refactor: :recycle: refactor and add more completion params to ui

* build: :building_construction: download from preview S3

* fix: :bug: fix paths

* fix: :green_heart: package:pre-release

* ci: :green_heart: more time for tests

* fix: :green_heart: fix build scripts

* fix: :bug: fix import in run.py

* fix: :bookmark: update version to try again

* ci: 💚 Update package.json version [skip ci]

* refactor: :fire: don't check for old extensions version

* fix: :bug: small bug fixes

* fix: :bug: fix config.py import paths

* ci: 💚 Update package.json version [skip ci]

* ci: :green_heart: platform-specific builds test #1

* feat: :green_heart: ship with binary

* fix: :green_heart: fix copy statement to include.exe for windows

* fix: :green_heart: cd extension before packaging

* chore: :loud_sound: count tokens generated

* fix: :green_heart: remove npm_config_arch

* fix: :green_heart: publish as pre-release!

* chore: :bookmark: update version

* perf: :green_heart: hardcode distro paths

* fix: :bug: fix yaml syntax error

* chore: :bookmark: update version

* fix: :green_heart: update permissions and version

* feat: :bug: kill old server if needed

* feat: :lipstick: update marketplace icon for pre-release

* ci: 💚 Update package.json version [skip ci]

* feat: :sparkles: auto-reload for config.py

* feat: :wrench: update default config.py imports

* feat: :sparkles: codelens in config.py

* feat: :sparkles: select model param count from UI

* ci: 💚 Update package.json version [skip ci]

* feat: :sparkles: more model options, ollama error handling

* perf: :zap: don't show server loading immediately

* fix: :bug: fixing small UI details

* ci: 💚 Update package.json version [skip ci]

* feat: :rocket: headers param on LLM class

* fix: :bug: fix headers for openai.;y

* feat: :sparkles: highlight code on cmd+shift+L

* ci: 💚 Update package.json version [skip ci]

* feat: :lipstick: sticky top bar in gui.tsx

* fix: :loud_sound: websocket logging and horizontal scrollbar

* ci: 💚 Update package.json version [skip ci]

* feat: :sparkles: allow AzureOpenAI Service through GGML

* ci: 💚 Update package.json version [skip ci]

* fix: :bug: fix automigration

* ci: 💚 Update package.json version [skip ci]

* ci: :green_heart: upload binaries in ci, download apple silicon

* chore: :fire: remove notes

* fix: :green_heart: use curl to download binary

* fix: :green_heart: set permissions on apple silicon binary

* fix: :green_heart: testing

* fix: :green_heart: cleanup file

* fix: :green_heart: fix preview.yaml

* fix: :green_heart: only upload once per binary

* fix: :green_heart: install rosetta

* ci: :green_heart: download binary after tests

* ci: 💚 Update package.json version [skip ci]

* ci: :green_heart: prepare ci for merge to main

---------

Co-authored-by: GitHub Action <action@github.com>

1 files changed, 78 insertions, 0 deletions

diff --git a/server/continuedev/libs/llm/hf_inference_api.py b/server/continuedev/libs/llm/hf_inference_api.py
new file mode 100644
index 00000000..990ec7c8
--- /dev/null
+++ b/server/continuedev/libs/llm/hf_inference_api.py
@@ -0,0 +1,78 @@
+from typing import Callable, Dict, List, Union
+
+from huggingface_hub import InferenceClient
+from pydantic import Field
+
+from .base import LLM, CompletionOptions
+from .prompts.chat import llama2_template_messages
+from .prompts.edit import simplified_edit_prompt
+
+
+class HuggingFaceInferenceAPI(LLM):
+    """
+    Hugging Face Inference API is a great option for newly released language models. Sign up for an account and add billing [here](https://huggingface.co/settings/billing), access the Inference Endpoints [here](https://ui.endpoints.huggingface.co), click on “New endpoint”, and fill out the form (e.g. select a model like [WizardCoder-Python-34B-V1.0](https://huggingface.co/WizardLM/WizardCoder-Python-34B-V1.0)), and then deploy your model by clicking “Create Endpoint”. Change `~/.continue/config.py` to look like this:
+
+    ```python title="~/.continue/config.py"
+    from continuedev.core.models import Models
+    from continuedev.libs.llm.hf_inference_api import HuggingFaceInferenceAPI
+
+    config = ContinueConfig(
+        ...
+        models=Models(
+            default=HuggingFaceInferenceAPI(
+                endpoint_url="<INFERENCE_API_ENDPOINT_URL>",
+                hf_token="<HUGGING_FACE_TOKEN>",
+        )
+    )
+    ```
+    """
+
+    model: str = Field(
+        "Hugging Face Inference API",
+        description="The name of the model to use (optional for the HuggingFaceInferenceAPI class)",
+    )
+    hf_token: str = Field(..., description="Your Hugging Face API token")
+    endpoint_url: str = Field(
+        None, description="Your Hugging Face Inference API endpoint URL"
+    )
+
+    template_messages: Union[
+        Callable[[List[Dict[str, str]]], str], None
+    ] = llama2_template_messages
+
+    prompt_templates = {
+        "edit": simplified_edit_prompt,
+    }
+
+    class Config:
+        arbitrary_types_allowed = True
+
+    def collect_args(self, options: CompletionOptions):
+        options.stop = None
+        args = super().collect_args(options)
+
+        if "max_tokens" in args:
+            args["max_new_tokens"] = args["max_tokens"]
+            del args["max_tokens"]
+        if "stop" in args:
+            args["stop_sequences"] = args["stop"]
+            del args["stop"]
+
+        return args
+
+    async def _stream_complete(self, prompt, options):
+        args = self.collect_args(options)
+
+        client = InferenceClient(self.endpoint_url, token=self.hf_token)
+
+        stream = client.text_generation(prompt, stream=True, details=True, **args)
+
+        for r in stream:
+            # skip special tokens
+            if r.token.special:
+                continue
+            # stop if we encounter a stop sequence
+            if options.stop is not None:
+                if r.token.text in options.stop:
+                    break
+            yield r.token.text