Merge pull request #34 from ServiceNow/intro-simplified

Simplified intro notebook
ServiceNow · Sep 28, 2024 · d2f842a · d2f842a
2 parents 814dbc4 + dab2172
commit d2f842a
Show file tree

Hide file tree

Showing 23 changed files with 1,264 additions and 1,672 deletions.
diff --git a/Makefile b/Makefile
@@ -1,4 +1,4 @@
-.PHONY: setup env install lint test test-slow test-all clean
+.PHONY: setup env install lint test test-slow test-all clean update-intro update-clean-intro clear-clean-intro
 
 ENV_NAME=tapeagents
 PYTHON_VERSION=3.10
@@ -37,5 +37,8 @@ update-intro:
 	cp examples/intro_clean.ipynb intro.ipynb
 	$(CONDA) run --name ${ENV_NAME} jupyter execute --inplace intro.ipynb
 
-clean-intro:
+update-clean-intro:
 	$(CONDA) run --name ${ENV_NAME} jupyter nbconvert intro.ipynb --output=examples/intro_clean.ipynb --to notebook --ClearOutputPreprocessor.enabled=True --ClearMetadataPreprocessor.enabled=True
+
+clear-clean-intro:
+	$(CONDA) run --name ${ENV_NAME} jupyter nbconvert --inplace examples/intro_clean.ipynb --ClearOutputPreprocessor.enabled=True --ClearMetadataPreprocessor.enabled=True
diff --git a/conf/tapeagent/llm/local_llama3_70b.yaml b/conf/tapeagent/llm/local_llama3_70b.yaml
@@ -1,7 +1,7 @@
-_target_: tapeagents.llms.LLAMA
+_target_: tapeagents.llms.TrainableLLM
 use_cache: false
 conf:
-  _target_: tapeagents.llms.LLAMAConfig
+  _target_: tapeagents.llms.TrainableLLMConfig
   base_url: http://0.0.0.0:8080
   model_name: meta-llama/Meta-Llama-3-70B-Instruct
   context_size: 8000

diff --git a/conf/tapeagent/llm/local_llama3_8b.yaml b/conf/tapeagent/llm/local_llama3_8b.yaml
@@ -1,7 +1,7 @@
-_target_: tapeagents.llms.LLAMA
+_target_: tapeagents.llms.TrainableLLM
 use_cache: false
 conf:
-  _target_: tapeagents.llms.LLAMAConfig
+  _target_: tapeagents.llms.TrainableLLMConfig
   base_url: http://0.0.0.0:8080
   model_name: meta-llama/Meta-Llama-3-8B-Instruct
   context_size: 8000

diff --git a/conf/tapeagent/llm/local_mixtral_8x22b.yaml b/conf/tapeagent/llm/local_mixtral_8x22b.yaml
@@ -1,7 +1,7 @@
-_target_: tapeagents.llms.LLAMA
+_target_: tapeagents.llms.TrainableLLM
 use_cache: false
 conf:
-  _target_: tapeagents.llms.LLAMAConfig
+  _target_: tapeagents.llms.TrainableLLMConfig
   base_url: http://0.0.0.0:8080
   model_name: mistralai/Mixtral-8x22B-v0.1
   context_size: 8000

diff --git a/example.env b/example.env
@@ -0,0 +1,8 @@
+# 1. Copy this file to your .env file that is gitignored to keep your keys safe
+# 2. In your .env only, uncomment and fill the following lines with your keys
+
+# OPENAI_API_KEY="" # put your https://platform.openai.com/ key here
+# OPENAI_ORGANIZATION="" # optional if you use your personal key
+
+# TOGETHER_API_KEY=""
+# HF_TOKEN=""
diff --git a/examples/agent.py b/examples/agent.py
@@ -3,7 +3,7 @@
 from tapeagents.agent import Agent, Node
 from tapeagents.core import SetNextNode, Prompt, Tape
 from tapeagents.dialog_tape import AssistantStep, AssistantThought, DialogTape, UserStep
-from tapeagents.llms import LLAMA, LLM, LLMStream
+from tapeagents.llms import TrainableLLM, LLM, LLMStream
 
 
 def hello_world(llm: LLM):
@@ -92,7 +92,7 @@ def generate_steps(self, agent, tape: Tape, llm_stream: LLMStream):
 
 
 if __name__ == "__main__":
-    llm = LLAMA(
+    llm = TrainableLLM(
         base_url="https://api.together.xyz",
         model_name="meta-llama/Meta-Llama-3-70B-Instruct-Turbo",
         tokenizer_name="meta-llama/Meta-Llama-3-70B-Instruct",

diff --git a/examples/batch_add_observations.py b/examples/batch_add_observations.py
@@ -4,7 +4,7 @@
 from tapeagents.batch import ObsLayerConfig, batch_add_observations
 from tapeagents.dialog_tape import AssistantStep, DialogTape, SystemStep, UserStep
 from tapeagents.io import save_tapes
-from tapeagents.llms import LLAMA, LLM
+from tapeagents.llms import TrainableLLM, LLM
 
 from .llama_user import LLAMAUserModel
 
@@ -40,7 +40,7 @@ def try_batch_add_observations(llm: LLM):
 
 if __name__ == "__main__":
     try_batch_add_observations(
-        LLAMA(
+        TrainableLLM(
             base_url="https://api.together.xyz",
             model_name="meta-llama/Llama-3-8b-chat-hf",
             parameters=dict(temperature=0.7, max_tokens=512),

diff --git a/examples/batch_main_loop.py b/examples/batch_main_loop.py
@@ -5,7 +5,7 @@
 from tapeagents.dialog_tape import DialogTape, SystemStep, UserStep
 from tapeagents.environment import EmptyEnvironment
 from tapeagents.io import save_tapes
-from tapeagents.llms import LLAMA, LLM
+from tapeagents.llms import TrainableLLM, LLM
 
 from .llama_agent import LLAMAChatBot
 
@@ -33,7 +33,7 @@ def try_batch_main_loop(llm: LLM):
 
 if __name__ == "__main__":
     try_batch_main_loop(
-        LLAMA(
+        TrainableLLM(
             base_url="https://api.together.xyz",
             model_name="meta-llama/Llama-3-8b-chat-hf",
             parameters=dict(temperature=0.7, max_tokens=512),

diff --git a/examples/chat.py b/examples/chat.py
@@ -2,7 +2,7 @@
 
 from tapeagents.team import TeamAgent, TeamTape
 from tapeagents.studio import Studio
-from tapeagents.llms import LLAMA, LLM
+from tapeagents.llms import TrainableLLM, LLM
 from tapeagents.rendering import PrettyRenderer
 
 
@@ -26,7 +26,7 @@ def try_chat(llm: LLM, studio: bool):
 
 
 if __name__ == "__main__":
-    llm = LLAMA(
+    llm = TrainableLLM(
         base_url="https://api.together.xyz",
         model_name="meta-llama/Meta-Llama-3-70B-Instruct-Turbo",
         parameters=dict(temperature=0.7, max_tokens=512),

diff --git a/examples/continue_tapes.py b/examples/continue_tapes.py
@@ -4,7 +4,7 @@
 from tapeagents.batch import generate_tapes
 from tapeagents.dialog_tape import AssistantStep, DialogTape, SystemStep, UserStep
 from tapeagents.environment import EmptyEnvironment
-from tapeagents.llms import LLAMA, LLM
+from tapeagents.llms import TrainableLLM, LLM
 
 from .llama_agent import LLAMAChatBot
 from .llama_user import LLAMAUserModel
@@ -48,7 +48,7 @@ def try_continue_tapes(llm: LLM):
 
 if __name__ == "__main__":
     try_continue_tapes(
-        LLAMA(
+        TrainableLLM(
             base_url="https://api.together.xyz",
             model_name="meta-llama/Llama-3-8b-chat-hf",
             parameters=dict(temperature=0.7, max_tokens=512),

diff --git a/examples/delegate.py b/examples/delegate.py
@@ -4,7 +4,7 @@
 
 from tapeagents.agent import Agent, AgentEvent
 from tapeagents.core import Action, Prompt, Tape, Thought
-from tapeagents.llms import LLAMA, LLM, LLMStream
+from tapeagents.llms import TrainableLLM, LLM, LLMStream
 
 logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s")
 
@@ -86,7 +86,7 @@ def delegate(self, tape: ExampleTape) -> Agent[ExampleTape]:
             raise ValueError(f"Invalid state {state}")
 
 
-def try_delegation(llama: LLAMA):
+def try_delegation(llama: TrainableLLM):
     tape = ExampleTape(context=EXAMPLE_TEXT)
     with open("start_tape.json", "w") as f:
         json.dump(tape.model_dump(), f, indent=2)
@@ -104,7 +104,7 @@ def try_delegation(llama: LLAMA):
 
 if __name__ == "__main__":
     try_delegation(
-        LLAMA(
+        TrainableLLM(
             base_url="https://api.together.xyz",
             model_name="meta-llama/Meta-Llama-3-70B-Instruct-Turbo",
             parameters=dict(temperature=0.7, max_tokens=512),

diff --git a/examples/delegate_stack.py b/examples/delegate_stack.py
@@ -14,7 +14,7 @@
     Tape,
     Thought,
 )
-from tapeagents.llms import LLAMA, LLM, LLMStream
+from tapeagents.llms import TrainableLLM, LLM, LLMStream
 from tapeagents.view import Call, Respond, TapeViewStack
 
 EXAMPLE_TEXT = """I am a text with some verbs like running, jumping, and swimming."""
@@ -224,7 +224,7 @@ def make_analyze_text_chain(llm: LLM):
 
 
 def main():
-    llama = LLAMA(
+    llama = TrainableLLM(
         base_url="https://api.together.xyz",
         model_name="meta-llama/Meta-Llama-3-70B-Instruct-Turbo",
         tokenizer_name="meta-llama/Meta-Llama-3-70B-Instruct",

diff --git a/examples/gsm8k_tuning/eval.ipynb b/examples/gsm8k_tuning/eval.ipynb
@@ -19,7 +19,7 @@
     "from termcolor import colored\n",
     "from tqdm import tqdm\n",
     "\n",
-    "from tapeagents.llms import LLAMA\n"
+    "from tapeagents.llms import TrainableLLM\n"
    ]
   },
   {
@@ -90,7 +90,7 @@
     "# run inference: vllm serve meta-llama/Meta-Llama-3.1-8B-Instruct\n",
     "untuned_agent = MathAgent(\n",
     "    llms={\n",
-    "        \"default\": LLAMA(\n",
+    "        \"default\": TrainableLLM(\n",
     "            base_url=\"http://localhost:8000\",\n",
     "            model_name=\"meta-llama/Meta-Llama-3.1-8B-Instruct\",\n",
     "            tokenizer_name=\"meta-llama/Meta-Llama-3.1-8B-Instruct\",\n",
@@ -415,7 +415,7 @@
     "# run inference: vllm serve gsm8k/tuning/llama31_70b_train_t02/tune1/intermediate/800/\n",
     "tuned_agent = MathAgent(\n",
     "    llms={\n",
-    "        \"default\": LLAMA(\n",
+    "        \"default\": TrainableLLM(\n",
     "            base_url=\"http://localhost:8000\",\n",
     "            model_name=\"gsm8k/tuning/llama31_70b_train_t02/tune1/intermediate/800/\",\n",
     "            tokenizer_name=\"meta-llama/Meta-Llama-3.1-8B-Instruct\",\n",
@@ -731,7 +731,7 @@
    "outputs": [],
    "source": [
     "# check teacher model\n",
-    "big_llm = LLAMA(\n",
+    "big_llm = TrainableLLM(\n",
     "    base_url=\"https://api.together.xyz\",\n",
     "    model_name=\"meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo\",\n",
     "    tokenizer_name=\"meta-llama/Meta-Llama-3.1-70B-Instruct\",\n",

diff --git a/examples/gsm8k_tuning/finetune.ipynb b/examples/gsm8k_tuning/finetune.ipynb
@@ -20,7 +20,7 @@
     "\n",
     "from tapeagents.core import TrainingText\n",
     "from math_agent import ActionExecutionFailure, MathAgent, MathTape\n",
-    "from tapeagents.llms import LLAMA\n"
+    "from tapeagents.llms import TrainableLLM\n"
    ]
   },
   {
@@ -32,7 +32,7 @@
     "# We need the agent to cut tapes into training samples\n",
     "agent = MathAgent(\n",
     "    llms={\n",
-    "        \"default\": LLAMA(\n",
+    "        \"default\": TrainableLLM(\n",
     "            base_url=\"https://api.together.xyz\",\n",
     "            model_name=\"meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo\",\n",
     "            tokenizer_name=\"meta-llama/Meta-Llama-3.1-8B-Instruct\",\n",

diff --git a/examples/gsm8k_tuning/produce_tapes.ipynb b/examples/gsm8k_tuning/produce_tapes.ipynb
@@ -20,7 +20,7 @@
     "from termcolor import colored\n",
     "from tqdm import tqdm\n",
     "\n",
-    "from tapeagents.llms import LLAMA\n"
+    "from tapeagents.llms import TrainableLLM\n"
    ]
   },
   {
@@ -52,7 +52,7 @@
     "np.random.seed(42)\n",
     "np.random.shuffle(samples)  # type: ignore\n",
     "\n",
-    "llm = LLAMA(\n",
+    "llm = TrainableLLM(\n",
     "    base_url=\"https://api.together.xyz\",\n",
     "    model_name=\"meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo\",\n",
     "    tokenizer_name=\"meta-llama/Meta-Llama-3.1-70B-Instruct\",\n",