Added Groq Support

docs/language-models/hosted-models/groq.mdx

@@ -0,0 +1,72 @@
+---
+title: Groq
+---
+
+To use Open Interpreter with a model from Groq, simply run:
+
+<CodeGroup>
+
+```bash Terminal
+interpreter --model groq/llama3-8b-8192
+```
+
+```python Python
+from interpreter import interpreter
+
+interpreter.llm.model = "groq/llama3-8b-8192"
+interpreter.llm.api_key = '<groq-api-key>'
+interpreter.chat()
+```
+
+</CodeGroup>
+
+If you are having any issues when passing the `--model`, try adding the `--api_base`:
+
+<CodeGroup>
+
+```bash Terminal
+interpreter --api_base "https://api.groq.com/openai/v1" --model groq/llama3-8b-8192 --api_key $GROQ_API_KEY
+```
+
+```python Python
+from interpreter import interpreter
+
+interpreter.llm.model = "groq/llama3-8b-8192"
+interpreter.llm.api_key = '<groq-api-key>'
+interpreter.llm.api_base = "https://api.groq.com/openai/v1"
+interpreter.llm.context_window = 32000
+interpreter.chat()
+```
+
+</CodeGroup>
+
+# Supported Models
+
+We support any model on [Groq's models page:](https://console.groq.com/docs/models)
+
+<CodeGroup>
+
+```bash Terminal
+interpreter --model groq/mixtral-8x7b-32768
+interpreter --model groq/llama3-8b-8192
+interpreter --model groq/llama3-70b-8192
+interpreter --model groq/gemma-7b-it
+```
+
+```python Python
+interpreter.llm.model = "groq/mixtral-8x7b-32768"
+interpreter.llm.model = "groq/llama3-8b-8192"
+interpreter.llm.model = "groq/llama3-70b-8192"
+interpreter.llm.model = "groq/gemma-7b-it"
+```
+
+</CodeGroup>
+
+# Required Environment Variables
+
+Run `export GROQ_API_KEY='<your-key-here>'` or place it in your rc file and re-source
+Set the following environment variables [(click here to learn how)](https://chat.openai.com/share/1062cdd8-62a1-4aa8-8ec9-eca45645971a) to use these models.
+
+| Environment Variable | Description | Where to Find |
+| -------------------- | ---------------------------------------------------- | ------------------------------------------------------------------- |
+| `GROQ_API_KEY` | The API key for authenticating to Groq's services. | [Groq Account Page](https://console.groq.com/keys) |

interpreter/core/llm/llm.py

@@ -1,4 +1,8 @@
 import litellm
+from groq import Groq
+
+groq_client = [None]
+
 import tokentrim as tt
 
 from ...terminal_interface.utils.display_markdown_message import (
@@ -9,6 +13,7 @@
 from .utils.convert_to_openai_messages import convert_to_openai_messages
 
 litellm.suppress_debug_info = True
+import os
 import time
 
 
@@ -26,6 +31,7 @@ def __init__(self, interpreter):
 
  # Settings
  self.model = "gpt-4-turbo"
+ self.model = "groq/mixtral-8x7b-32768" # can now use models from groq. `export GROQ_API_KEY="your-key-here")` or use --model
  self.temperature = 0
  self.supports_vision = False
  self.supports_functions = None # Will try to auto-detect
@@ -67,7 +73,7 @@ def run(self, messages):
  self.supports_functions = False
  except:
  self.supports_functions = False
- 
+
  # Trim image messages if they're there
  if self.supports_vision:
  image_messages = [msg for msg in messages if msg["type"] == "image"]
@@ -210,7 +216,54 @@ def fixed_litellm_completions(**params):
  # Run completion
  first_error = None
  try:
- yield from litellm.completion(**params)
+ def source(**params):
+ """Get Completions Using LiteLLM"""
+ yield from litellm.completion(**params)
+
+ if "model" in params and "groq/" in params["model"]:
+
+ def groq_complete(**params):
+ if groq_client[0] is None:
+ groq_client[0] = Groq(
+ # This is the default and can be omitted
+ api_key=os.environ.get("GROQ_API_KEY"),
+ timeout=2,
+ max_retries=3,
+ )
+ res = (
+ groq_client[0]
+ .chat.completions.create(
+ messages=params["messages"],
+ model=params["model"].split("groq/")[1],
+ )
+ .choices[0]
+ .message.content
+ )
+ return res
+
+ def s(**params):
+ """Get Completions Using Groq"""
+ params["stream"] = (
+ False # To keep things simple for now, and groq is super fast anyway
+ )
+ word_by_word = False
+ if word_by_word:
+ for chunk in groq_complete(**params).split(" "):
+ yield {
+ "choices": [
+ {"delta": {"type": "message", "content": chunk + " "}}
+ ]
+ }
+ else:
+ for whole in [groq_complete(**params)]:
+ yield {
+ "choices": [
+ {"delta": {"type": "message", "content": whole}}
+ ]
+ }
+
+ source = s
+ yield from source(**params)
  except Exception as e:
  # Store the first error
  first_error = e

pyproject.toml

@@ -29,6 +29,7 @@ matplotlib = "^3.8.2"
 toml = "^0.10.2"
 posthog = "^3.1.0"
 tiktoken = "^0.6.0"
+groq = "^4.3.0"
 
 
 #Optional dependencies