synth-ai 0.2.8.dev2__py3-none-any.whl → 0.4.3__py3-none-any.whl

synth_ai/cli/rl_demo.py

@@ -1,202 +0,0 @@
-#!/usr/bin/env python3
-"""
-New RL demo command group kept fully separate from legacy demo.
-
-Usage examples:
-  uvx synth-ai rl_demo check
-  uvx synth-ai rl_demo deploy --app /path/to/math_task_app.py --name synth-math-demo
-  uvx synth-ai rl_demo configure
-  uvx synth-ai rl_demo run --batch-size 4 --group-size 16 --model Qwen/Qwen3-0.6B
-  uvx synth-ai run --config demo_config.toml
-
-For convenience, dotted aliases are also exposed:
-  uvx synth-ai rl_demo.check
-"""
-
-from __future__ import annotations
-
-import click
-
-
-def _forward(args: list[str]) -> None:
-    import sys
-    try:
-        from synth_ai.demos.core import cli as demo_cli  # type: ignore
-    except Exception as e:  # pragma: no cover
-        click.echo(f"Failed to import RL demo CLI: {e}")
-        sys.exit(1)
-    rc = int(demo_cli.main(args) or 0)
-    if rc != 0:
-        sys.exit(rc)
-
-
-def register(cli):
-    @cli.group("rl_demo")
-    def rl_demo():
-        """RL Demo commands (separate from legacy demo)."""
-
-    # Help pyright understand dynamic Click group attributes
-    from typing import Any, cast as _cast
-    _rlg = _cast(Any, rl_demo)
-
-    @_rlg.command("setup")
-    def rl_setup():
-        _forward(["rl_demo.setup"])  # primary setup command
-
-    # (prepare command removed; consolidated into configure)
-
-    @_rlg.command("deploy")
-    @click.option("--local", is_flag=True, help="Run local FastAPI instead of Modal deploy")
-    @click.option("--app", type=click.Path(), default=None, help="Path to Modal app.py for uv run modal deploy")
-    @click.option("--name", type=str, default="synth-math-demo", help="Modal app name")
-    @click.option("--script", type=click.Path(), default=None, help="Path to deploy_task_app.sh (optional legacy)")
-    def rl_deploy(local: bool, app: str | None, name: str, script: str | None):
-        args: list[str] = ["rl_demo.deploy"]
-        if local:
-            args.append("--local")
-        if app:
-            args.extend(["--app", app])
-        if name:
-            args.extend(["--name", name])
-        if script:
-            args.extend(["--script", script])
-        _forward(args)
-
-    @_rlg.command("configure")
-    def rl_configure():
-        _forward(["rl_demo.configure"]) 
-
-    @_rlg.command("init")
-    @click.option("--force", is_flag=True, help="Overwrite existing files in CWD")
-    def rl_init(force: bool):
-        args = ["rl_demo.init"]
-        if force:
-            args.append("--force")
-        _forward(args)
-
-    @_rlg.command("run")
-    @click.option("--config", type=click.Path(), default=None, help="Path to TOML config (skip prompt)")
-    @click.option("--batch-size", type=int, default=None)
-    @click.option("--group-size", type=int, default=None)
-    @click.option("--model", type=str, default=None)
-    @click.option("--timeout", type=int, default=600)
-    @click.option("--dry-run", is_flag=True, help="Print request body and exit")
-    def rl_run(config: str | None, batch_size: int | None, group_size: int | None, model: str | None, timeout: int, dry_run: bool):
-        args = ["rl_demo.run"]
-        if config:
-            args.extend(["--config", config])
-        if batch_size is not None:
-            args.extend(["--batch-size", str(batch_size)])
-        if group_size is not None:
-            args.extend(["--group-size", str(group_size)])
-        if model:
-            args.extend(["--model", model])
-        if timeout is not None:
-            args.extend(["--timeout", str(timeout)])
-        if dry_run:
-            args.append("--dry-run")
-        _forward(args)
-
-    # Dotted aliases (top-level): legacy check → setup
-    @cli.command("rl_demo.check")
-    def rl_check_alias():
-        _forward(["rl_demo.setup"]) 
-
-    @cli.command("rl_demo.setup")
-    def rl_setup_alias():
-        _forward(["rl_demo.setup"]) 
-
-    # (prepare alias removed)
-
-    @cli.command("rl_demo.deploy")
-    @click.option("--local", is_flag=True, help="Run local FastAPI instead of Modal deploy")
-    @click.option("--app", type=click.Path(), default=None, help="Path to Modal app.py for uv run modal deploy")
-    @click.option("--name", type=str, default="synth-math-demo", help="Modal app name")
-    @click.option("--script", type=click.Path(), default=None, help="Path to deploy_task_app.sh (optional legacy)")
-    def rl_deploy_alias(local: bool, app: str | None, name: str, script: str | None):
-        args: list[str] = ["rl_demo.deploy"]
-        if local:
-            args.append("--local")
-        if app:
-            args.extend(["--app", app])
-        if name:
-            args.extend(["--name", name])
-        if script:
-            args.extend(["--script", script])
-        _forward(args)
-
-    @cli.command("rl_demo.configure")
-    def rl_configure_alias():
-        _forward(["rl_demo.configure"]) 
-
-    @cli.command("rl_demo.init")
-    @click.option("--force", is_flag=True, help="Overwrite existing files in CWD")
-    def rl_init_alias(force: bool):
-        args = ["rl_demo.init"]
-        if force:
-            args.append("--force")
-        _forward(args)
-
-    @cli.command("rl_demo.run")
-    @click.option("--config", type=click.Path(), default=None, help="Path to TOML config (skip prompt)")
-    @click.option("--batch-size", type=int, default=None)
-    @click.option("--group-size", type=int, default=None)
-    @click.option("--model", type=str, default=None)
-    @click.option("--timeout", type=int, default=600)
-    @click.option("--dry-run", is_flag=True, help="Print request body and exit")
-    def rl_run_alias(config: str | None, batch_size: int | None, group_size: int | None, model: str | None, timeout: int, dry_run: bool):
-        args = ["rl_demo.run"]
-        if config:
-            args.extend(["--config", config])
-        if batch_size is not None:
-            args.extend(["--batch-size", str(batch_size)])
-        if group_size is not None:
-            args.extend(["--group-size", str(group_size)])
-        if model:
-            args.extend(["--model", model])
-        if timeout is not None:
-            args.extend(["--timeout", str(timeout)])
-        if dry_run:
-            args.append("--dry-run")
-        _forward(args)
-
-    # Top-level convenience alias: `synth-ai deploy`
-    @cli.command("deploy")
-    @click.option("--local", is_flag=True, help="Run local FastAPI instead of Modal deploy")
-    @click.option("--app", type=click.Path(), default=None, help="Path to Modal app.py for uv run modal deploy")
-    @click.option("--name", type=str, default="synth-math-demo", help="Modal app name")
-    @click.option("--script", type=click.Path(), default=None, help="Path to deploy_task_app.sh (optional legacy)")
-    def deploy_top(local: bool, app: str | None, name: str, script: str | None):
-        args: list[str] = ["rl_demo.deploy"]
-        if local:
-            args.append("--local")
-        if app:
-            args.extend(["--app", app])
-        if name:
-            args.extend(["--name", name])
-        if script:
-            args.extend(["--script", script])
-        _forward(args)
-
-    @cli.command("run")
-    @click.option("--config", type=click.Path(), default=None, help="Path to TOML config (skip prompt)")
-    @click.option("--batch-size", type=int, default=None)
-    @click.option("--group-size", type=int, default=None)
-    @click.option("--model", type=str, default=None)
-    @click.option("--timeout", type=int, default=600)
-    @click.option("--dry-run", is_flag=True, help="Print request body and exit")
-    def run_top(config: str | None, batch_size: int | None, group_size: int | None, model: str | None, timeout: int, dry_run: bool):
-        args = ["run"]
-        if config:
-            args.extend(["--config", config])
-        if batch_size is not None:
-            args.extend(["--batch-size", str(batch_size)])
-        if group_size is not None:
-            args.extend(["--group-size", str(group_size)])
-        if model:
-            args.extend(["--model", model])
-        if timeout is not None:
-            args.extend(["--timeout", str(timeout)])
-        if dry_run:
-            args.append("--dry-run")
-        _forward(args)

synth_ai/cli/status.py

@@ -1,133 +0,0 @@
-#!/usr/bin/env python3
-"""
-CLI: status of agent runs/versions and environment service.
-"""
-
-import asyncio
-
-import click
-import requests
-from rich import box
-from rich.console import Console
-from rich.panel import Panel
-from rich.table import Table
-
-
-async def _db_stats(db_url: str) -> dict:
-    from synth_ai.tracing_v3.turso.manager import AsyncSQLTraceManager
-
-    db = AsyncSQLTraceManager(db_url)
-    await db.initialize()
-    try:
-        out: dict = {}
-        # Totals
-        totals = await db.query_traces(
-            """
-            SELECT 
-              (SELECT COUNT(*) FROM session_traces) AS sessions,
-              (SELECT COUNT(*) FROM experiments) AS experiments,
-              (SELECT COUNT(*) FROM events) AS events,
-              (SELECT COUNT(*) FROM messages) AS messages,
-              (SELECT COALESCE(SUM(CASE WHEN event_type='cais' THEN cost_usd ELSE 0 END),0)/100.0 FROM events) AS total_cost_usd,
-              (SELECT COALESCE(SUM(CASE WHEN event_type='cais' THEN total_tokens ELSE 0 END),0) FROM events) AS total_tokens
-        """
-        )
-        if not totals.empty:
-            out["totals"] = totals.iloc[0].to_dict()
-        else:
-            out["totals"] = {}
-
-        # Systems summary
-        systems = await db.query_traces(
-            """
-            SELECT system_type, COUNT(*) as count FROM systems GROUP BY system_type
-        """
-        )
-        out["systems"] = systems
-
-        versions = await db.query_traces(
-            """
-            SELECT COUNT(*) as version_count FROM system_versions
-        """
-        )
-        if not versions.empty:
-            out["version_count"] = int(versions.iloc[0]["version_count"])
-        else:
-            out["version_count"] = 0
-        return out
-    finally:
-        await db.close()
-
-
-def register(cli):
-    @cli.command()
-    @click.option(
-        "--url",
-        "db_url",
-        default="sqlite+aiosqlite:///./synth_ai.db/dbs/default/data",
-        help="Database URL",
-    )
-    @click.option("--service-url", default="http://127.0.0.1:8901", help="Environment service URL")
-    def status(db_url: str, service_url: str):
-        """Show DB stats, agent/environment system counts, and env service health."""
-        console = Console()
-
-        async def _run():
-            # DB
-            stats = await _db_stats(db_url)
-
-            # Env service
-            health_text = "[red]unreachable[/red]"
-            envs_list = []
-            try:
-                r = requests.get(f"{service_url}/health", timeout=2)
-                if r.ok:
-                    data = r.json()
-                    health_text = "[green]ok[/green]"
-                    envs_list = data.get("supported_environments", [])
-                else:
-                    health_text = f"[red]{r.status_code}[/red]"
-            except Exception:
-                pass
-
-            # Render
-            totals = stats.get("totals", {})
-            lines = []
-            lines.append(f"DB: [dim]{db_url}[/dim]")
-            lines.append(
-                f"Experiments: {int(totals.get('experiments', 0)):,}  "
-                f"Sessions: {int(totals.get('sessions', 0)):,}  "
-                f"Events: {int(totals.get('events', 0)):,}  "
-                f"Messages: {int(totals.get('messages', 0)):,}"
-            )
-            lines.append(
-                f"Cost: ${float(totals.get('total_cost_usd', 0.0) or 0.0):.4f}  "
-                f"Tokens: {int(totals.get('total_tokens', 0)):,}"
-            )
-            lines.append("")
-            lines.append(f"Env Service: {health_text}  [dim]{service_url}[/dim]")
-            if envs_list:
-                lines.append(
-                    "Environments: "
-                    + ", ".join(sorted(envs_list)[:10])
-                    + (" ..." if len(envs_list) > 10 else "")
-                )
-
-            panel_main = Panel("\n".join(lines), title="Synth AI Status", border_style="cyan")
-            console.print(panel_main)
-
-            # Systems table
-            sys_df = stats.get("systems")
-            if sys_df is not None and not sys_df.empty:
-                tbl = Table(
-                    title=f"Systems (versions: {stats.get('version_count', 0)})",
-                    box=box.SIMPLE,
-                    header_style="bold",
-                )
-                tbl.add_column("Type")
-                tbl.add_column("Count", justify="right")
-                for _, r in sys_df.iterrows():
-                    tbl.add_row(str(r.get("system_type", "-")), f"{int(r.get('count', 0)):,}")
-                console.print(tbl)
-
-        asyncio.run(_run())

synth_ai/config/base_url.py

@@ -1,107 +0,0 @@
-"""
-Base URL resolution for learning-v2 and related backend APIs.
-
-Default to production, allow overrides via environment variables:
-- LEARNING_V2_BASE_URL (highest precedence)
-- SYNTH_BASE_URL (legacy)
-- SYNTH_LOCAL_BASE_URL
-- SYNTH_DEV_BASE_URL
-- SYNTH_PROD_BASE_URL (fallback if none provided)
-
-Normalization: ensure the returned URL ends with "/api".
-"""
-
-import os
-from typing import Literal
-
-PROD_BASE_URL_DEFAULT = "https://agent-learning.onrender.com"
-
-
-def _normalize_base(url: str) -> str:
-    url = url.strip()
-    if url.endswith("/v1"):
-        url = url[:-3]
-    url = url.rstrip("/")
-    if not url.endswith("/api"):
-        url = f"{url}/api"
-    return url
-
-
-def get_learning_v2_base_url(mode: Literal["dev", "prod"] = "prod") -> str:
-    if mode == "prod":
-        prod = os.getenv("SYNTH_PROD_BASE_URL") or PROD_BASE_URL_DEFAULT
-        return _normalize_base(prod)
-    # Priority order
-    env_url = os.getenv("LEARNING_V2_BASE_URL")
-    if env_url:
-        return _normalize_base(env_url)
-
-    legacy = os.getenv("SYNTH_BASE_URL")
-    if legacy:
-        return _normalize_base(legacy)
-
-    local = os.getenv("SYNTH_LOCAL_BASE_URL")
-    if local:
-        return _normalize_base(local)
-
-    dev = os.getenv("SYNTH_DEV_BASE_URL")
-    if dev:
-        return _normalize_base(dev)
-
-    raise Exception()
-
-
-def _resolve_override_mode() -> str:
-    """Return one of 'local', 'dev', 'prod' based on SYNTH_BACKEND_URL_OVERRIDE.
-
-    Defaults to 'prod' when unset or unrecognized.
-    """
-    ov = (os.getenv("SYNTH_BACKEND_URL_OVERRIDE", "") or "").strip().lower()
-    if ov in {"local", "dev", "prod"}:
-        return ov
-    return "prod"
-
-
-def get_backend_from_env() -> tuple[str, str]:
-    """Resolve (base_url, api_key) using a simple LOCAL/DEV/PROD override scheme.
-
-    Env vars consulted:
-    - BACKEND_OVERRIDE = full URL (with or without /api)
-    - SYNTH_BACKEND_URL_OVERRIDE = local|dev|prod (case-insensitive)
-    - LOCAL_BACKEND_URL, TESTING_LOCAL_SYNTH_API_KEY
-    - DEV_BACKEND_URL, DEV_SYNTH_API_KEY
-    - PROD_BACKEND_URL, TESTING_PROD_SYNTH_API_KEY (fallback to SYNTH_API_KEY)
-
-    Base URL is normalized to end with '/api'.
-    Defaults: prod base URL → https://agent-learning.onrender.com/api
-    """
-    direct_override = (os.getenv("BACKEND_OVERRIDE") or "").strip()
-    if direct_override:
-        base = direct_override.rstrip("/")
-        if base.endswith("/api"):
-            base = base[: -len("/api")]
-        api_key = os.getenv("SYNTH_API_KEY", "").strip()
-        return base, api_key
-
-    mode = _resolve_override_mode()
-    if mode == "local":
-        base = os.getenv("LOCAL_BACKEND_URL", "http://localhost:8000")
-        key = os.getenv("TESTING_LOCAL_SYNTH_API_KEY", "")
-        return base.rstrip("/"), key
-    if mode == "dev":
-        base = os.getenv("DEV_BACKEND_URL", "") or "http://localhost:8000"
-        key = os.getenv("DEV_SYNTH_API_KEY", "")
-        return base.rstrip("/"), key
-    # prod
-    base = os.getenv("PROD_BACKEND_URL", f"{PROD_BASE_URL_DEFAULT}")
-    # Ensure we return the root (no trailing /api). If default includes /api, strip it.
-    base = base.rstrip("/")
-    if base.endswith("/api"):
-        base = base[: -len("/api")]
-    # Prefer explicit PROD key, then testing key, then generic fallback
-    key = (
-        os.getenv("PROD_SYNTH_API_KEY", "")
-        or os.getenv("TESTING_PROD_SYNTH_API_KEY", "")
-        or os.getenv("SYNTH_API_KEY", "")
-    )
-    return base, key

synth_ai/core/experiment.py

@@ -1,15 +0,0 @@
-
-
-class ExperimentalSystem:
-    system_id: str
-    system_version_id: str
-    pass
-
-
-class Experiment:
-    id: str
-    name: str
-    description: str
-    created_at: str
-    updated_at: str
-    related_systems: list[ExperimentalSystem]

synth_ai/core/system.py

@@ -1,15 +0,0 @@
-class System:
-    id: str
-    name: str
-    description: str
-    pass
-
-
-class SystemVersion:
-    id: str
-    system_id: str
-    branch: str
-    commit: str
-    created_at: str
-    description: str
-    pass

synth_ai/demos/core/__init__.py

@@ -1 +0,0 @@
-# Core demo CLI and helpers package

synth_ai/demos/demo_task_apps/__init__.py

@@ -1 +0,0 @@
-# Namespace for demo task apps (math, etc.)

synth_ai/demos/demo_task_apps/math/config.toml

@@ -1,129 +0,0 @@
-[model]
-name = "Qwen/Qwen3-0.6B"
-dtype = "bfloat16"
-seed = 42
-trainer_mode = "full"
-
-[lora]
-r = 16
-alpha = 32
-dropout = 0.05
-target_modules = [
-  "q_proj", "k_proj", "v_proj", "o_proj",
-  "gate_proj", "up_proj", "down_proj",
-]
-
-[rdma]
-enabled = false
-ifname = "eth0"
-ip_type = "ipv4"
-p2p_disable = 0
-shm_disable = 0
-fast_nccl = false
-
-gid_index = 3
-cross_nic = 0
-collnet_enable = 0
-net_gdr_level = 2
-
-nsocks_perthread = 4
-socket_nthreads = 2
-
-algo = "Ring"
-proto = "Simple"
-p2p_level = "SYS"
-debug = "INFO"
-
-[reference]
-placement = "dedicated"
-gpu_index = 1
-port = 8002
-tp = 1
-health_max_wait_s = 180
-health_interval_ms = 300
-
-[topology]
-type = "single_node_split"
-gpu_type = "H100:4"
-use_rdma = false
-gpus_for_vllm = 2
-gpus_for_training = 1
-gpus_for_ref = 1
-tensor_parallel = 2
-
-[training]
-num_epochs = 1
-iterations_per_epoch = 2
-batch_size = 1
-group_size = 8
-learning_rate = 5e-6
-max_grad_norm = 0.5
-log_interval = 1
-update_reference_interval = 0
-weight_sync_interval = 1
-
-[training.weight_sync]
-enable = true
-targets = ["policy"]
-
-[vllm]
-tensor_parallel_size = 2
-gpu_memory_utilization = 0.9
-max_model_len = 8192
-max_num_seqs = 32
-enforce_eager = false
-max_parallel_generations = 4
-
-[evaluation]
-seeds = [0, 1, 2, 3, 4, 5, 6, 7]
-rollouts_per_seed = 1
-instances = 0
-max_concurrent_rollouts = 4
-thinking_mode = "think"
-every_n_iters = 5
-
-[rollout]
-env_name = "math"
-policy_name = "math-react"
-env_config = {}
-max_steps_per_episode = 5
-sampling_temperature = 0.3
-sampling_top_p = 0.95
-max_tokens = 1024
-max_concurrent_rollouts = 4
-ops_per_rollout = 14
-on_done = "reset"
-thinking_mode = "think"
-thinking_budget = 512
-
-[policy]
-config = {}
-
-[hyperparams]
-epsilon_low = 0.1
-epsilon_high = 0.3
-delta = 5.0
-beta = 0.01
-kl_penalty = 0.01
-advantage_normalization = true
-group_normalization = true
-num_inner_steps = 1
-clip_epsilon = 0.2
-completion_only = false
-
-[step_rewards]
-enabled = false
-mode = "off"
-step_beta = 0.0
-indicator_lambda = 0.0
-
-[trainer]
-allow_ref_fallback = false
-
-[checkpoint]
-interval = 10
-directory = "/checkpoints"
-keep_last_n = 3
-save_optimizer = true
-save_scheduler = true
-enabled = true

synth_ai/demos/demo_task_apps/math/deploy_task_app.sh

@@ -1,22 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-log_file="${script_dir}/.last_deploy.log"
-rm -f "$log_file"
-
-: "${ENVIRONMENT_API_KEY:=}"
-
-repo_root="$(cd "${script_dir}/../../../.." && pwd)"
-fallback_script="${repo_root}/examples/rl/deploy_task_app.sh"
-
-if [[ -f "${fallback_script}" ]]; then
-  echo "Using ${fallback_script} via 'uv run'" | tee -a "$log_file"
-  (cd "${repo_root}/examples/rl" && ENVIRONMENT_API_KEY="${ENVIRONMENT_API_KEY}" uv run bash "${fallback_script}" | tee -a "$log_file")
-else
-  echo "ERROR: Deploy script not found at ${fallback_script}. Pass --script /path/to/deploy_task_app.sh" | tee -a "$log_file"
-  exit 1
-fi
-
-echo "Deploy finished. Inspect $log_file for the public URL (…modal.run)." | tee -a "$log_file"
-