Spaces:
Running
Running
Update agents.py
Browse files
agents.py
CHANGED
@@ -3,31 +3,26 @@ import json
|
|
3 |
import asyncio
|
4 |
import random
|
5 |
|
6 |
-
|
7 |
-
import os
|
8 |
-
import json
|
9 |
-
import asyncio
|
10 |
-
import random
|
11 |
-
|
12 |
# --- OpenAI ---
|
13 |
-
from openai import AsyncOpenAI, APIError
|
14 |
|
15 |
# --- Google Gemini ---
|
16 |
from google import genai
|
17 |
from google.genai import types
|
18 |
-
|
19 |
-
|
20 |
|
21 |
# --- Mistral AI ---
|
22 |
-
from mistralai.async_client import MistralAsyncClient
|
|
|
23 |
|
24 |
# --- Poke-Env ---
|
25 |
from poke_env.player import Player
|
26 |
from poke_env.environment.battle import Battle
|
27 |
from poke_env.environment.move import Move
|
28 |
from poke_env.environment.pokemon import Pokemon
|
|
|
29 |
|
30 |
-
# --- Helper Function & Base Class
|
31 |
def normalize_name(name: str) -> str:
|
32 |
"""Lowercase and remove non-alphanumeric characters."""
|
33 |
return "".join(filter(str.isalnum, name)).lower()
|
@@ -64,14 +59,13 @@ STANDARD_TOOL_SCHEMA = {
|
|
64 |
}
|
65 |
|
66 |
|
67 |
-
class LLMAgentBase(Player):
|
68 |
def __init__(self, *args, **kwargs):
|
69 |
super().__init__(*args, **kwargs)
|
70 |
self.standard_tools = STANDARD_TOOL_SCHEMA
|
71 |
-
self.battle_history = []
|
72 |
|
73 |
def _format_battle_state(self, battle: Battle) -> str:
|
74 |
-
# (Implementation as provided in the question)
|
75 |
active_pkmn = battle.active_pokemon
|
76 |
active_pkmn_info = f"Your active Pokemon: {active_pkmn.species} " \
|
77 |
f"(Type: {'/'.join(map(str, active_pkmn.types))}) " \
|
@@ -117,9 +111,7 @@ class LLMAgentBase(Player): # Make sure this base class exists
|
|
117 |
f"Opponent Side Conditions: {battle.opponent_side_conditions}"
|
118 |
return state_str.strip()
|
119 |
|
120 |
-
|
121 |
-
def _find_move_by_name(self, battle: Battle, move_name: str) -> Move | None:
|
122 |
-
# (Implementation as provided in the question)
|
123 |
normalized_name = normalize_name(move_name)
|
124 |
# Prioritize exact ID match
|
125 |
for move in battle.available_moves:
|
@@ -132,8 +124,7 @@ class LLMAgentBase(Player): # Make sure this base class exists
|
|
132 |
return move
|
133 |
return None
|
134 |
|
135 |
-
def _find_pokemon_by_name(self, battle: Battle, pokemon_name: str) -> Pokemon
|
136 |
-
# (Implementation as provided in the question)
|
137 |
normalized_name = normalize_name(pokemon_name)
|
138 |
for pkmn in battle.available_switches:
|
139 |
# Normalize the species name for comparison
|
@@ -142,7 +133,6 @@ class LLMAgentBase(Player): # Make sure this base class exists
|
|
142 |
return None
|
143 |
|
144 |
async def choose_move(self, battle: Battle) -> str:
|
145 |
-
# (Implementation as provided in the question - relies on _get_llm_decision)
|
146 |
battle_state_str = self._format_battle_state(battle)
|
147 |
decision_result = await self._get_llm_decision(battle_state_str)
|
148 |
decision = decision_result.get("decision")
|
@@ -160,8 +150,7 @@ class LLMAgentBase(Player): # Make sure this base class exists
|
|
160 |
if chosen_move and chosen_move in battle.available_moves:
|
161 |
action_taken = True
|
162 |
chat_msg = f"AI Decision: Using move '{chosen_move.id}'."
|
163 |
-
print(chat_msg)
|
164 |
-
# await self.send_message(chat_msg, battle=battle) # Uncomment if send_message exists
|
165 |
return self.create_order(chosen_move)
|
166 |
else:
|
167 |
fallback_reason = f"LLM chose unavailable/invalid move '{move_name}'."
|
@@ -174,8 +163,7 @@ class LLMAgentBase(Player): # Make sure this base class exists
|
|
174 |
if chosen_switch and chosen_switch in battle.available_switches:
|
175 |
action_taken = True
|
176 |
chat_msg = f"AI Decision: Switching to '{chosen_switch.species}'."
|
177 |
-
print(chat_msg)
|
178 |
-
# await self.send_message(chat_msg, battle=battle) # Uncomment if send_message exists
|
179 |
return self.create_order(chosen_switch)
|
180 |
else:
|
181 |
fallback_reason = f"LLM chose unavailable/invalid switch '{pokemon_name}'."
|
@@ -185,55 +173,53 @@ class LLMAgentBase(Player): # Make sure this base class exists
|
|
185 |
fallback_reason = f"LLM called unknown function '{function_name}'."
|
186 |
|
187 |
if not action_taken:
|
188 |
-
if not fallback_reason:
|
189 |
if error_message:
|
190 |
fallback_reason = f"API Error: {error_message}"
|
191 |
-
elif decision is None:
|
192 |
fallback_reason = "LLM did not provide a valid function call."
|
193 |
-
else:
|
194 |
fallback_reason = "Unknown error processing LLM decision."
|
195 |
|
196 |
print(f"Warning: {fallback_reason} Choosing random action.")
|
197 |
-
# await self.send_message(f"AI Fallback: {fallback_reason} Choosing random action.", battle=battle) # Uncomment
|
198 |
|
199 |
-
# Use poke-env's built-in random choice
|
200 |
if battle.available_moves or battle.available_switches:
|
201 |
return self.choose_random_move(battle)
|
202 |
else:
|
203 |
print("AI Fallback: No moves or switches available. Using Struggle/Default.")
|
204 |
-
|
205 |
-
return self.choose_default_move(battle) # Handles struggle
|
206 |
|
207 |
-
async def _get_llm_decision(self, battle_state: str) ->
|
208 |
raise NotImplementedError("Subclasses must implement _get_llm_decision")
|
209 |
|
|
|
210 |
# --- Google Gemini Agent ---
|
211 |
class GeminiAgent(LLMAgentBase):
|
212 |
"""Uses Google Gemini API for decisions."""
|
213 |
-
def __init__(self, api_key: str
|
214 |
super().__init__(*args, **kwargs)
|
215 |
self.model_name = model
|
216 |
used_api_key = api_key or os.environ.get("GOOGLE_API_KEY")
|
217 |
-
self.model_name=model
|
218 |
if not used_api_key:
|
219 |
raise ValueError("Google API key not provided or found in GOOGLE_API_KEY env var.")
|
220 |
-
|
221 |
-
|
222 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
223 |
)
|
224 |
-
# --- Correct Tool Definition ---
|
225 |
-
# Create a list of function declaration dictionaries from the values in STANDARD_TOOL_SCHEMA
|
226 |
-
function_declarations = list(self.standard_tools.values())
|
227 |
-
# Create the Tool object expected by the API
|
228 |
-
self.gemini_tool_config = types.Tool(function_declarations=function_declarations)
|
229 |
-
# --- End Tool Definition ---
|
230 |
-
|
231 |
-
# --- Correct Model Initialization ---
|
232 |
-
# Pass the Tool object directly to the model's 'tools' parameter
|
233 |
|
234 |
-
|
235 |
-
|
236 |
-
async def _get_llm_decision(self, battle_state: str) -> dict:
|
237 |
"""Sends state to the Gemini API and gets back the function call decision."""
|
238 |
prompt = (
|
239 |
"You are a skilled Pokemon battle AI. Your goal is to win the battle. "
|
@@ -246,63 +232,64 @@ class GeminiAgent(LLMAgentBase):
|
|
246 |
)
|
247 |
|
248 |
try:
|
249 |
-
#
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
model=self.model_name,
|
254 |
-
contents=prompt
|
255 |
)
|
256 |
-
|
257 |
-
|
258 |
-
# --- Response Parsing (Your logic was already good here) ---
|
259 |
-
# Check candidates and parts safely
|
260 |
if not response.candidates:
|
261 |
-
|
262 |
-
|
263 |
-
|
264 |
-
|
|
|
|
|
265 |
|
266 |
candidate = response.candidates[0]
|
267 |
if not candidate.content or not candidate.content.parts:
|
268 |
finish_reason_str = "Unknown"
|
269 |
-
try:
|
270 |
-
|
|
|
|
|
271 |
return {"error": f"Gemini response issue. Finish Reason: {finish_reason_str}"}
|
272 |
|
273 |
-
part
|
274 |
-
|
275 |
-
|
276 |
-
|
277 |
-
|
278 |
-
|
279 |
-
|
280 |
-
|
281 |
-
|
282 |
-
|
283 |
-
|
284 |
-
|
285 |
-
|
286 |
-
|
287 |
-
|
288 |
-
|
289 |
-
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
294 |
-
|
295 |
-
|
|
|
296 |
except Exception as e:
|
297 |
-
# Catch any other unexpected errors during the API call or processing
|
298 |
print(f"Unexpected error during Gemini processing: {e}")
|
299 |
import traceback
|
300 |
-
traceback.print_exc()
|
301 |
return {"error": f"Unexpected error: {str(e)}"}
|
|
|
|
|
302 |
# --- OpenAI Agent ---
|
303 |
class OpenAIAgent(LLMAgentBase):
|
304 |
"""Uses OpenAI API for decisions."""
|
305 |
-
def __init__(self, api_key: str
|
306 |
super().__init__(*args, **kwargs)
|
307 |
self.model = model
|
308 |
used_api_key = api_key or os.environ.get("OPENAI_API_KEY")
|
@@ -311,9 +298,9 @@ class OpenAIAgent(LLMAgentBase):
|
|
311 |
self.openai_client = AsyncOpenAI(api_key=used_api_key)
|
312 |
|
313 |
# Convert standard schema to OpenAI's format
|
314 |
-
self.
|
315 |
|
316 |
-
async def _get_llm_decision(self, battle_state: str) ->
|
317 |
system_prompt = (
|
318 |
"You are a skilled Pokemon battle AI. Your goal is to win the battle. "
|
319 |
"Based on the current battle state, decide the best action: either use an available move or switch to an available Pokémon. "
|
@@ -330,25 +317,26 @@ class OpenAIAgent(LLMAgentBase):
|
|
330 |
{"role": "system", "content": system_prompt},
|
331 |
{"role": "user", "content": user_prompt},
|
332 |
],
|
333 |
-
|
334 |
-
|
335 |
temperature=0.5,
|
336 |
)
|
337 |
message = response.choices[0].message
|
338 |
-
|
339 |
-
|
340 |
-
|
|
|
|
|
341 |
try:
|
342 |
-
|
343 |
-
|
344 |
-
|
345 |
-
return {"decision": {"name": function_name, "arguments": arguments}}
|
346 |
else:
|
347 |
-
|
348 |
except json.JSONDecodeError:
|
349 |
-
return {"error": f"Error decoding function
|
350 |
else:
|
351 |
-
# Model decided not to call a function
|
352 |
return {"error": f"OpenAI did not return a function call. Response: {message.content}"}
|
353 |
|
354 |
except APIError as e:
|
@@ -362,7 +350,7 @@ class OpenAIAgent(LLMAgentBase):
|
|
362 |
# --- Mistral Agent ---
|
363 |
class MistralAgent(LLMAgentBase):
|
364 |
"""Uses Mistral AI API for decisions."""
|
365 |
-
def __init__(self, api_key: str
|
366 |
super().__init__(*args, **kwargs)
|
367 |
self.model = model
|
368 |
used_api_key = api_key or os.environ.get("MISTRAL_API_KEY")
|
@@ -370,51 +358,51 @@ class MistralAgent(LLMAgentBase):
|
|
370 |
raise ValueError("Mistral API key not provided or found in MISTRAL_API_KEY env var.")
|
371 |
self.mistral_client = MistralAsyncClient(api_key=used_api_key)
|
372 |
|
373 |
-
# Convert standard schema to Mistral's tool format
|
374 |
-
self.mistral_tools =
|
375 |
|
376 |
-
async def _get_llm_decision(self, battle_state: str) ->
|
377 |
system_prompt = (
|
378 |
"You are a skilled Pokemon battle AI. Your goal is to win the battle. "
|
379 |
"Based on the current battle state, decide the best action: either use an available move or switch to an available Pokémon. "
|
380 |
"Consider type matchups, HP, status conditions, field effects, entry hazards, and potential opponent actions. "
|
381 |
"Only choose actions listed as available using their exact ID (for moves) or species name (for switches). "
|
382 |
-
"Use the provided tools
|
383 |
)
|
384 |
user_prompt = f"Current Battle State:\n{battle_state}\n\nChoose the best action by calling the appropriate function ('choose_move' or 'choose_switch')."
|
385 |
|
386 |
try:
|
387 |
-
response = await self.mistral_client.chat
|
388 |
model=self.model,
|
389 |
messages=[
|
390 |
-
{"role": "system", "content":
|
391 |
-
{"role": "user", "content":
|
392 |
],
|
393 |
tools=self.mistral_tools,
|
394 |
-
tool_choice="auto",
|
395 |
temperature=0.5,
|
396 |
)
|
397 |
|
398 |
message = response.choices[0].message
|
399 |
-
#
|
400 |
if message.tool_calls:
|
401 |
-
tool_call =
|
402 |
function_name = tool_call.function.name
|
403 |
-
|
404 |
-
|
405 |
-
|
406 |
-
|
407 |
-
|
|
|
|
|
|
|
408 |
else:
|
409 |
-
# Model decided not to call a tool
|
410 |
-
|
411 |
|
412 |
-
|
413 |
-
|
414 |
-
|
|
|
415 |
print(f"Error during Mistral API call: {e}")
|
416 |
-
|
417 |
-
error_details = str(e)
|
418 |
-
# if isinstance(e, MistralAPIException): # Example
|
419 |
-
# error_details = f"{e.status_code} - {e.message}"
|
420 |
-
return {"error": f"Mistral API Error: {error_details}"}
|
|
|
3 |
import asyncio
|
4 |
import random
|
5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
# --- OpenAI ---
|
7 |
+
from openai import AsyncOpenAI, APIError
|
8 |
|
9 |
# --- Google Gemini ---
|
10 |
from google import genai
|
11 |
from google.genai import types
|
12 |
+
from google.api_core import exceptions as google_exceptions
|
|
|
13 |
|
14 |
# --- Mistral AI ---
|
15 |
+
from mistralai.async_client import MistralAsyncClient
|
16 |
+
from mistralai.exceptions import MistralAPIError
|
17 |
|
18 |
# --- Poke-Env ---
|
19 |
from poke_env.player import Player
|
20 |
from poke_env.environment.battle import Battle
|
21 |
from poke_env.environment.move import Move
|
22 |
from poke_env.environment.pokemon import Pokemon
|
23 |
+
from typing import Optional, Dict, Any, Union
|
24 |
|
25 |
+
# --- Helper Function & Base Class ---
|
26 |
def normalize_name(name: str) -> str:
|
27 |
"""Lowercase and remove non-alphanumeric characters."""
|
28 |
return "".join(filter(str.isalnum, name)).lower()
|
|
|
59 |
}
|
60 |
|
61 |
|
62 |
+
class LLMAgentBase(Player):
|
63 |
def __init__(self, *args, **kwargs):
|
64 |
super().__init__(*args, **kwargs)
|
65 |
self.standard_tools = STANDARD_TOOL_SCHEMA
|
66 |
+
self.battle_history = []
|
67 |
|
68 |
def _format_battle_state(self, battle: Battle) -> str:
|
|
|
69 |
active_pkmn = battle.active_pokemon
|
70 |
active_pkmn_info = f"Your active Pokemon: {active_pkmn.species} " \
|
71 |
f"(Type: {'/'.join(map(str, active_pkmn.types))}) " \
|
|
|
111 |
f"Opponent Side Conditions: {battle.opponent_side_conditions}"
|
112 |
return state_str.strip()
|
113 |
|
114 |
+
def _find_move_by_name(self, battle: Battle, move_name: str) -> Optional[Move]:
|
|
|
|
|
115 |
normalized_name = normalize_name(move_name)
|
116 |
# Prioritize exact ID match
|
117 |
for move in battle.available_moves:
|
|
|
124 |
return move
|
125 |
return None
|
126 |
|
127 |
+
def _find_pokemon_by_name(self, battle: Battle, pokemon_name: str) -> Optional[Pokemon]:
|
|
|
128 |
normalized_name = normalize_name(pokemon_name)
|
129 |
for pkmn in battle.available_switches:
|
130 |
# Normalize the species name for comparison
|
|
|
133 |
return None
|
134 |
|
135 |
async def choose_move(self, battle: Battle) -> str:
|
|
|
136 |
battle_state_str = self._format_battle_state(battle)
|
137 |
decision_result = await self._get_llm_decision(battle_state_str)
|
138 |
decision = decision_result.get("decision")
|
|
|
150 |
if chosen_move and chosen_move in battle.available_moves:
|
151 |
action_taken = True
|
152 |
chat_msg = f"AI Decision: Using move '{chosen_move.id}'."
|
153 |
+
print(chat_msg)
|
|
|
154 |
return self.create_order(chosen_move)
|
155 |
else:
|
156 |
fallback_reason = f"LLM chose unavailable/invalid move '{move_name}'."
|
|
|
163 |
if chosen_switch and chosen_switch in battle.available_switches:
|
164 |
action_taken = True
|
165 |
chat_msg = f"AI Decision: Switching to '{chosen_switch.species}'."
|
166 |
+
print(chat_msg)
|
|
|
167 |
return self.create_order(chosen_switch)
|
168 |
else:
|
169 |
fallback_reason = f"LLM chose unavailable/invalid switch '{pokemon_name}'."
|
|
|
173 |
fallback_reason = f"LLM called unknown function '{function_name}'."
|
174 |
|
175 |
if not action_taken:
|
176 |
+
if not fallback_reason:
|
177 |
if error_message:
|
178 |
fallback_reason = f"API Error: {error_message}"
|
179 |
+
elif decision is None:
|
180 |
fallback_reason = "LLM did not provide a valid function call."
|
181 |
+
else:
|
182 |
fallback_reason = "Unknown error processing LLM decision."
|
183 |
|
184 |
print(f"Warning: {fallback_reason} Choosing random action.")
|
|
|
185 |
|
|
|
186 |
if battle.available_moves or battle.available_switches:
|
187 |
return self.choose_random_move(battle)
|
188 |
else:
|
189 |
print("AI Fallback: No moves or switches available. Using Struggle/Default.")
|
190 |
+
return self.choose_default_move(battle)
|
|
|
191 |
|
192 |
+
async def _get_llm_decision(self, battle_state: str) -> Dict[str, Any]:
|
193 |
raise NotImplementedError("Subclasses must implement _get_llm_decision")
|
194 |
|
195 |
+
|
196 |
# --- Google Gemini Agent ---
|
197 |
class GeminiAgent(LLMAgentBase):
|
198 |
"""Uses Google Gemini API for decisions."""
|
199 |
+
def __init__(self, api_key: str = None, model: str = "gemini-1.5-flash", *args, **kwargs):
|
200 |
super().__init__(*args, **kwargs)
|
201 |
self.model_name = model
|
202 |
used_api_key = api_key or os.environ.get("GOOGLE_API_KEY")
|
|
|
203 |
if not used_api_key:
|
204 |
raise ValueError("Google API key not provided or found in GOOGLE_API_KEY env var.")
|
205 |
+
|
206 |
+
# Initialize Gemini client
|
207 |
+
genai.configure(api_key=used_api_key)
|
208 |
+
|
209 |
+
# Configure the model with tools
|
210 |
+
self.gemini_tool_config = [
|
211 |
+
{
|
212 |
+
"function_declarations": list(self.standard_tools.values())
|
213 |
+
}
|
214 |
+
]
|
215 |
+
|
216 |
+
# Initialize the model
|
217 |
+
self.model = genai.GenerativeModel(
|
218 |
+
model_name=self.model_name,
|
219 |
+
tools=self.gemini_tool_config
|
220 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
221 |
|
222 |
+
async def _get_llm_decision(self, battle_state: str) -> Dict[str, Any]:
|
|
|
|
|
223 |
"""Sends state to the Gemini API and gets back the function call decision."""
|
224 |
prompt = (
|
225 |
"You are a skilled Pokemon battle AI. Your goal is to win the battle. "
|
|
|
232 |
)
|
233 |
|
234 |
try:
|
235 |
+
# Use the async API for Gemini
|
236 |
+
response = await self.model.generate_content_async(
|
237 |
+
prompt,
|
238 |
+
generation_config={"temperature": 0.5}
|
|
|
|
|
239 |
)
|
240 |
+
|
|
|
|
|
|
|
241 |
if not response.candidates:
|
242 |
+
finish_reason_str = "No candidates found"
|
243 |
+
try:
|
244 |
+
finish_reason_str = response.prompt_feedback.block_reason.name
|
245 |
+
except AttributeError:
|
246 |
+
pass
|
247 |
+
return {"error": f"Gemini response issue. Reason: {finish_reason_str}"}
|
248 |
|
249 |
candidate = response.candidates[0]
|
250 |
if not candidate.content or not candidate.content.parts:
|
251 |
finish_reason_str = "Unknown"
|
252 |
+
try:
|
253 |
+
finish_reason_str = candidate.finish_reason.name
|
254 |
+
except AttributeError:
|
255 |
+
pass
|
256 |
return {"error": f"Gemini response issue. Finish Reason: {finish_reason_str}"}
|
257 |
|
258 |
+
for part in candidate.content.parts:
|
259 |
+
if hasattr(part, 'function_call') and part.function_call:
|
260 |
+
fc = part.function_call
|
261 |
+
function_name = fc.name
|
262 |
+
# Convert arguments to dict
|
263 |
+
arguments = {}
|
264 |
+
if fc.args:
|
265 |
+
arguments = {k: v for k, v in fc.args.items()}
|
266 |
+
|
267 |
+
if function_name in self.standard_tools:
|
268 |
+
return {"decision": {"name": function_name, "arguments": arguments}}
|
269 |
+
else:
|
270 |
+
return {"error": f"Model called unknown function '{function_name}'. Args: {arguments}"}
|
271 |
+
|
272 |
+
# If we got here, no function call was found in any part
|
273 |
+
text_content = " ".join([
|
274 |
+
part.text if hasattr(part, 'text') else str(part)
|
275 |
+
for part in candidate.content.parts
|
276 |
+
])
|
277 |
+
return {"error": f"Gemini did not return a function call. Response: {text_content[:100]}..."}
|
278 |
+
|
279 |
+
except google_exceptions.GoogleAPIError as e:
|
280 |
+
print(f"Google API error: {e}")
|
281 |
+
return {"error": f"Google API error: {str(e)}"}
|
282 |
except Exception as e:
|
|
|
283 |
print(f"Unexpected error during Gemini processing: {e}")
|
284 |
import traceback
|
285 |
+
traceback.print_exc()
|
286 |
return {"error": f"Unexpected error: {str(e)}"}
|
287 |
+
|
288 |
+
|
289 |
# --- OpenAI Agent ---
|
290 |
class OpenAIAgent(LLMAgentBase):
|
291 |
"""Uses OpenAI API for decisions."""
|
292 |
+
def __init__(self, api_key: str = None, model: str = "gpt-4o", *args, **kwargs):
|
293 |
super().__init__(*args, **kwargs)
|
294 |
self.model = model
|
295 |
used_api_key = api_key or os.environ.get("OPENAI_API_KEY")
|
|
|
298 |
self.openai_client = AsyncOpenAI(api_key=used_api_key)
|
299 |
|
300 |
# Convert standard schema to OpenAI's format
|
301 |
+
self.openai_tools = list(self.standard_tools.values())
|
302 |
|
303 |
+
async def _get_llm_decision(self, battle_state: str) -> Dict[str, Any]:
|
304 |
system_prompt = (
|
305 |
"You are a skilled Pokemon battle AI. Your goal is to win the battle. "
|
306 |
"Based on the current battle state, decide the best action: either use an available move or switch to an available Pokémon. "
|
|
|
317 |
{"role": "system", "content": system_prompt},
|
318 |
{"role": "user", "content": user_prompt},
|
319 |
],
|
320 |
+
tools=self.openai_tools,
|
321 |
+
tool_choice="auto", # Let the model choose
|
322 |
temperature=0.5,
|
323 |
)
|
324 |
message = response.choices[0].message
|
325 |
+
|
326 |
+
# Check for tool calls in the response
|
327 |
+
if message.tool_calls:
|
328 |
+
tool_call = message.tool_calls[0] # Get the first tool call
|
329 |
+
function_name = tool_call.function.name
|
330 |
try:
|
331 |
+
arguments = json.loads(tool_call.function.arguments or '{}')
|
332 |
+
if function_name in self.standard_tools:
|
333 |
+
return {"decision": {"name": function_name, "arguments": arguments}}
|
|
|
334 |
else:
|
335 |
+
return {"error": f"Model called unknown function '{function_name}'."}
|
336 |
except json.JSONDecodeError:
|
337 |
+
return {"error": f"Error decoding function arguments: {tool_call.function.arguments}"}
|
338 |
else:
|
339 |
+
# Model decided not to call a function
|
340 |
return {"error": f"OpenAI did not return a function call. Response: {message.content}"}
|
341 |
|
342 |
except APIError as e:
|
|
|
350 |
# --- Mistral Agent ---
|
351 |
class MistralAgent(LLMAgentBase):
|
352 |
"""Uses Mistral AI API for decisions."""
|
353 |
+
def __init__(self, api_key: str = None, model: str = "mistral-large-latest", *args, **kwargs):
|
354 |
super().__init__(*args, **kwargs)
|
355 |
self.model = model
|
356 |
used_api_key = api_key or os.environ.get("MISTRAL_API_KEY")
|
|
|
358 |
raise ValueError("Mistral API key not provided or found in MISTRAL_API_KEY env var.")
|
359 |
self.mistral_client = MistralAsyncClient(api_key=used_api_key)
|
360 |
|
361 |
+
# Convert standard schema to Mistral's tool format
|
362 |
+
self.mistral_tools = list(self.standard_tools.values())
|
363 |
|
364 |
+
async def _get_llm_decision(self, battle_state: str) -> Dict[str, Any]:
|
365 |
system_prompt = (
|
366 |
"You are a skilled Pokemon battle AI. Your goal is to win the battle. "
|
367 |
"Based on the current battle state, decide the best action: either use an available move or switch to an available Pokémon. "
|
368 |
"Consider type matchups, HP, status conditions, field effects, entry hazards, and potential opponent actions. "
|
369 |
"Only choose actions listed as available using their exact ID (for moves) or species name (for switches). "
|
370 |
+
"Use the provided tools to indicate your choice."
|
371 |
)
|
372 |
user_prompt = f"Current Battle State:\n{battle_state}\n\nChoose the best action by calling the appropriate function ('choose_move' or 'choose_switch')."
|
373 |
|
374 |
try:
|
375 |
+
response = await self.mistral_client.chat(
|
376 |
model=self.model,
|
377 |
messages=[
|
378 |
+
{"role": "system", "content": system_prompt},
|
379 |
+
{"role": "user", "content": user_prompt}
|
380 |
],
|
381 |
tools=self.mistral_tools,
|
382 |
+
tool_choice="auto", # Let the model choose
|
383 |
temperature=0.5,
|
384 |
)
|
385 |
|
386 |
message = response.choices[0].message
|
387 |
+
# Check for tool calls in the response
|
388 |
if message.tool_calls:
|
389 |
+
tool_call = message.tool_calls[0] # Get the first tool call
|
390 |
function_name = tool_call.function.name
|
391 |
+
try:
|
392 |
+
arguments = json.loads(tool_call.function.arguments or '{}')
|
393 |
+
if function_name in self.standard_tools:
|
394 |
+
return {"decision": {"name": function_name, "arguments": arguments}}
|
395 |
+
else:
|
396 |
+
return {"error": f"Model called unknown function '{function_name}'."}
|
397 |
+
except json.JSONDecodeError:
|
398 |
+
return {"error": f"Error decoding function arguments: {tool_call.function.arguments}"}
|
399 |
else:
|
400 |
+
# Model decided not to call a tool
|
401 |
+
return {"error": f"Mistral did not return a tool call. Response: {message.content}"}
|
402 |
|
403 |
+
except MistralAPIError as e:
|
404 |
+
print(f"Error during Mistral API call: {e}")
|
405 |
+
return {"error": f"Mistral API Error: {str(e)}"}
|
406 |
+
except Exception as e:
|
407 |
print(f"Error during Mistral API call: {e}")
|
408 |
+
return {"error": f"Unexpected error: {str(e)}"}
|
|
|
|
|
|
|
|