tinyllama-cli

API Reference

This document provides complete reference for all CLI commands, scripts, and Python APIs.

CLI Commands

download_model.py

Downloads models from HuggingFace Hub.

python download_model.py [OPTIONS]

Options

Option Description Example
--model KEY Model key or custom model ID --model tinyllama

Model Keys

Exit Codes

Code Description
0 Success
1 Error (invalid model, download failed, etc.)

ai_cli.py

The main chat CLI.

python ai_cli.py [OPTIONS]

Options

Option Description Example
--model MODEL Model folder name, path, or ‘auto’ --model tinyllama

Examples

# Auto-select model based on query
python ai_cli.py --model auto

# Use specific model
python ai_cli.py --model TinyLlama-1.1B-Chat-v1.0

# Use local path
python ai_cli.py --model ./models/my-model/

tinyllama.sh

Bootstrap script for automated setup.

./tinyllama.sh [OPTIONS]

Options

Option Description
--bootstrap-only Download deps and model, don’t start CLI
--model MODEL Auto-download specific model

Examples

# Full bootstrap + launch
./tinyllama.sh

# Download only
./tinyllama.sh --bootstrap-only

# Download specific model
./tinyllama.sh --model nvidia_nemotron

Python API

TinyLlamaCLI

Main chat interface class.

from ai_cli import TinyLlamaCLI
from pathlib import Path

# Initialize
cli = TinyLlamaCLI(
    model_dir=Path("models/TinyLlama-1.1B-Chat-v1.0"),
    model_label="TinyLlama"
)

# Run the CLI
cli.run()

Constructor Parameters

Parameter Type Description
model_dir Path Path to model directory
model_label str Display label for the model

Methods

run()

Starts the interactive chat loop.

cli.run()
_generate_response(user_input: str) -> str

Generates a response to user input.

response = cli._generate_response("Hello!")
_save_transcript() -> None

Saves the current chat transcript.

cli._save_transcript()
_export_training_data() -> None

Exports training data in JSONL format.

cli._export_training_data()

GenerationConfig

Configuration for text generation.

from ai_cli import GenerationConfig

cfg = GenerationConfig(
    temperature=0.65,
    top_p=0.9,
    top_k=40,
    repetition_penalty=1.1,
    max_new_tokens=256,
    do_sample=True
)

Parameters

Parameter Type Default Description
temperature float 0.65 Sampling temperature (0=deterministic)
top_p float 0.9 Nucleus sampling threshold
top_k int 40 Top-k sampling
repetition_penalty float 1.1 Repetition penalty
max_new_tokens int 256 Maximum tokens to generate
do_sample bool True Use sampling vs greedy

TinyLlamaOptimizer

Automatic tuning for generation settings.

from ai_cli import TinyLlamaOptimizer

cfg = TinyLlamaOptimizer.tune(
    user_input="Explain Python decorators",
    turns=1
)

Methods

tune(user_input: str, turns: int) -> GenerationConfig

Automatically tunes settings based on input.

cfg = TinyLlamaOptimizer.tune("Write a poem", 0)

download_model Functions

MODEL_CHOICES

Dictionary of pre-configured models.

from download_model import MODEL_CHOICES

print(MODEL_CHOICES)
# {'tinyllama': 'TinyLlama/TinyLlama-1.1B-Chat-v1.0', ...}

model_dir_for(model_id: str) -> Path

Get the local directory for a model.

from download_model import model_dir_for

path = model_dir_for("TinyLlama/TinyLlama-1.1B-Chat-v1.0")
# Returns: Path('models/TinyLlama-1.1B-Chat-v1.0')

Module Functions

parse_args()

Parse command line arguments.

from ai_cli import parse_args

args = parse_args()
# args.model contains the --model value

Constants

DEFAULT_MODEL_ID

from ai_cli import DEFAULT_MODEL_ID
# "TinyLlama/TinyLlama-1.1B-Chat-v1.0"

DEFAULT_MODEL_DIR

from ai_cli import DEFAULT_MODEL_DIR
# Path("models/TinyLlama-1.1B-Chat-v1.0")

SYSTEM_PROMPT

from ai_cli import SYSTEM_PROMPT
# "You are a helpful, concise AI assistant..."

Web Search API

search_web(query: str) -> list[WebResult]

Search the web for information.

from web_search import search_web

results = search_web("Python decorators tutorial")
for result in results:
    print(result.title, result.url)

Returns

List of WebResult objects:

@dataclass
class WebResult:
    title: str
    url: str
    snippet: str

should_search_web(query: str) -> bool

Determine if a query needs web search.

from web_search import should_search_web

if should_search_web("latest AI news"):
    # Search the web
    pass

Data Structures

ChatMessage

@dataclass
class ChatMessage:
    role: str      # "system", "user", "assistant"
    content: str
    timestamp: str  # ISO 8601

Transcript

@dataclass
class Transcript:
    model: str
    model_path: str
    started_at: str
    settings: dict
    messages: list[ChatMessage]

TrainingDataRecord

@dataclass
class TrainingDataRecord:
    id: str
    source_transcript: str
    created_at: str
    messages: list[dict]  # [{"role": "...", "content": "..."}]

Next Steps

This site is open source. Improve this page.