Upload folder using huggingface_hub

.dockerignore

@@ -0,0 +1,21 @@
+.git
+.venv
+__pycache__/
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+
+data/generated/
+data/district_llm_dataset_v1/
+data/district_llm_dataset_v2/
+data/district_llm_dataset_v3/
+
+artifacts/district_llm_adapter_v2/
+artifacts/dqn_shared/checkpoints/
+artifacts/dqn_shared/tensorboard/
+artifacts/llm_runtime_diagnosis/
+artifacts/quick_rl_llm_eval/
+artifacts/rl_guidance_eval/
+artifacts/rl_llm_wrapper_sweep/
+
+notebooks/

.gitattributes

@@ -33,3 +33,19 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v2/checkpoint-100/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v2/checkpoint-150/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v2/checkpoint-200/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v2/checkpoint-250/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v2/checkpoint-300/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v2/checkpoint-50/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v2/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v3/main_run/adapter/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v3/main_run/checkpoints/checkpoint-100/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v3/main_run/checkpoints/checkpoint-150/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v3/main_run/checkpoints/checkpoint-200/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+artifacts/district_llm_adapter_v3/main_run/checkpoints/checkpoint-50/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+third_party/CityFlow/examples/replay.txt filter=lfs diff=lfs merge=lfs -text
+third_party/CityFlow/extern/pybind11/.git.bak/objects/pack/pack-0b2353194187af7e228cfabd32025bb4d3af8551.idx filter=lfs diff=lfs merge=lfs -text
+third_party/CityFlow/extern/pybind11/.git.bak/objects/pack/pack-0b2353194187af7e228cfabd32025bb4d3af8551.pack filter=lfs diff=lfs merge=lfs -text
+third_party/CityFlow/extern/pybind11/.git.bak/objects/pack/pack-0b2353194187af7e228cfabd32025bb4d3af8551.rev filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,18 @@
+__pycache__
+*.pyc
+data/generated/*
+.DS_Store
+data/
+build/
+build-docker/
+local/
+.vs/
+.vscode/
+.idea/
+.DS_Store
+__pycache__
+CMakeSettings.json
+cmake-build-*
+CityFlow.egg-info
+frontend/replay/*
+notebooks/unsloth_compiled_cache

Dockerfile.openenv-api

@@ -0,0 +1,82 @@
+# OpenEnv API container
+#
+# Two-stage build:
+#   1. builder  - compiles the vendored CityFlow Python extension
+#   2. runtime  - installs the API dependencies and copies the repo-local data
+#
+# Runtime env vars:
+#   DATA_DIR         generated CityFlow dataset root
+#                    default: /app/data/generated
+#   SPLITS_DIR       train/val/test split files
+#                    default: /app/data/splits
+#   CHECKPOINT_PATH  optional DQN checkpoint
+#                    default: /app/artifacts/dqn_shared/best_validation.pt
+# ---------------------------------------------------------------------------
+
+# ── Stage 1: Build CityFlow ─────────────────────────────────────────────────
+FROM python:3.12-slim AS builder
+
+# Build tools needed by CityFlow's CMake build
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        build-essential \
+        cmake \
+        libboost-all-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /build
+
+# Copy only the CityFlow source (pybind11 is bundled as an extern submodule)
+COPY third_party/CityFlow ./CityFlow
+
+# Build and install cityflow into the builder's site-packages
+RUN rm -rf ./CityFlow/build
+RUN pip install --no-cache-dir ./CityFlow
+
+# Locate the compiled .so so we can copy it to the runtime stage
+RUN python -c "import cityflow, os; print(os.path.dirname(cityflow.__file__))"
+
+
+# ── Stage 2: Runtime ────────────────────────────────────────────────────────
+FROM python:3.12-slim AS runtime
+
+WORKDIR /app
+
+# Copy the compiled cityflow extension from the builder
+COPY --from=builder /usr/local/lib/python3.12/site-packages/cityflow* \
+                    /usr/local/lib/python3.12/site-packages/
+
+# Install Python dependencies (no build tools needed here)
+COPY openenv_app/requirements.txt ./requirements.txt
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy application source (only what the OpenEnv API needs at runtime)
+COPY agents/         ./agents/
+COPY district_llm/   ./district_llm/
+COPY env/            ./env/
+COPY openenv_app/    ./openenv_app/
+COPY server/__init__.py           ./server/__init__.py
+COPY server/path_validators.py    ./server/path_validators.py
+COPY server/policy_runner.py      ./server/policy_runner.py
+COPY server/roadnet_matcher.py    ./server/roadnet_matcher.py
+COPY training/       ./training/
+COPY data/splits/    ./data/splits/
+COPY data/generated/city_0002/ ./data/generated/city_0002/
+COPY artifacts/dqn_shared/best_validation.pt ./artifacts/dqn_shared/best_validation.pt
+COPY artifacts/district_llm_adapter_v3/main_run/adapter/ ./artifacts/district_llm_adapter_v3/main_run/adapter/
+
+# Keep runtime paths present, but expect the actual generated dataset to be
+# mounted or synced separately instead of baked into the image.
+RUN mkdir -p /app/data/generated /app/data/splits
+
+# Default paths (overridable at runtime via env vars).
+# DATA_DIR must point at a mounted/generated dataset root that contains city_*/
+# directories and scenario files; only data/splits is bundled here.
+ENV DATA_DIR=/app/data/generated
+ENV SPLITS_DIR=/app/data/splits
+ENV CHECKPOINT_PATH=/app/artifacts/dqn_shared/best_validation.pt
+ENV DISTRICT_LLM_ADAPTER_PATH=/app/artifacts/district_llm_adapter_v3/main_run/adapter
+
+# OpenEnv and HF Spaces commonly use port 7860.
+EXPOSE 7860
+
+CMD ["sh", "-c", "uvicorn openenv_app.app:app --host 0.0.0.0 --port ${PORT:-7860}"]

Dockerfile.visualizer

@@ -0,0 +1,41 @@
+# ── HF Space 2: Visualizer Dashboard ──────────────────────────────────────
+#
+# Lightweight image — no CityFlow C++ build, no torch.
+# All simulation runs are delegated to Space 1 (OpenEnv API) via HTTP.
+#
+# HF Spaces expects the app to listen on port 7860.
+#
+# Required env vars (set in the Space settings or README front matter):
+#   OPENENV_API_URL   URL of the OpenEnv API Space
+#                     e.g. https://your-org-openenv-api.hf.space
+#
+# Optional:
+#   DATA_DIR          city config root for roadnet matching (default: /app/data/bundled)
+#   REPLAY_ROOT       where replays are cached on disk (default: /app/results/replays)
+# ---------------------------------------------------------------------------
+
+FROM python:3.12-slim
+
+WORKDIR /app
+
+# Install dependencies (no cmake / build-essential needed)
+COPY server/requirements.txt ./requirements.txt
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Application source
+COPY server/           ./server/
+COPY third_party/CityFlow/frontend/ ./third_party/CityFlow/frontend/
+
+# Bundled city data (for roadnet matching / city/scenario dropdowns)
+COPY data/bundled/  ./data/bundled/
+COPY data/splits/   ./data/splits/
+
+# Writable directory for cached replays
+RUN mkdir -p /app/results/replays
+
+ENV DATA_DIR=/app/data/bundled
+ENV REPLAY_ROOT=/app/results/replays
+
+EXPOSE 7860
+
+CMD ["uvicorn", "server.visualizer_app:app", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -0,0 +1,199 @@
+---
+title: Agentic Traffic
+emoji: 🏢
+colorFrom: green
+colorTo: purple
+sdk: docker
+pinned: false
+short_description: Agentic AI to control traffic lights
+app_port: 7860
+---
+
+# traffic-llm
+
+CityFlow-based traffic-control project with intersection-level multi-agent DQN training and district-aware policy variants.
+
+## OpenEnv UI
+
+For the deployed OpenEnv web interface:
+
+- Click `Reset` before using `Step`.
+- Leave `Use Llm` unchecked for the fast, stable DQN-only path.
+- Use `District Actions` = `{}` for a valid no-op step payload.
+- Only enable `Use Llm` when you explicitly want district-level LLM guidance on top of the DQN executor.
+
+## Training
+
+The default local-policy trainer now uses parameter-shared dueling Double DQN with prioritized replay and n-step returns:
+
+```bash
+python3 -m training.train_local_policy train
+```
+
+That trains against `data/generated`, uses `data/splits`, writes checkpoints to `artifacts/dqn_shared`, enables TensorBoard logging, uses parallel CPU rollout workers by default, shows `tqdm` progress bars, and now validates plus checkpoints every 40 updates by default.
+
+For a broader but still manageable validation pass:
+
+```bash
+python3 -m training.train_local_policy train --max-val-cities 3 --val-scenarios-per-city 7
+```
+
+That evaluates 3 validation cities across all 7 scenario types. This gives 21 learned-policy validation episodes per eval, or 63 total episodes if random and fixed baselines are also enabled.
+
+Phase-3-style full training with the same 40-update eval/checkpoint cadence:
+
+```bash
+python3 -m training.train_local_policy train \
+  --max-train-cities 70 \
+  --max-val-cities 3 \
+  --val-scenarios-per-city 7 \
+  --policy-arch single_head_with_district_feature \
+  --reward-variant wait_queue_throughput
+```
+
+Useful ablations:
+
+```bash
+python3 -m training.train_local_policy train --policy-arch multi_head --reward-variant current
+python3 -m training.train_local_policy train --policy-arch single_head --reward-variant current
+python3 -m training.train_local_policy train --policy-arch single_head_with_district_feature --reward-variant wait_queue_throughput
+```
+
+For a fast phase-1 overfit run on one fixed world:
+
+```bash
+python3 -m training.train_local_policy train \
+  --total-updates 25 \
+  --train-city-id city_0072 \
+  --train-scenario-name normal \
+  --overfit-val-on-train-scenario \
+  --fast-overfit \
+  --policy-arch single_head_with_district_feature \
+  --reward-variant wait_queue_throughput
+```
+
+To create or refresh dataset splits:
+
+```bash
+python3 -m training.train_local_policy make-splits
+```
+
+To evaluate the best checkpoint:
+
+```bash
+python3 -m training.train_local_policy evaluate \
+  --checkpoint artifacts/dqn_shared/best_validation.pt \
+  --split val
+```
+
+To evaluate a heuristic baseline directly:
+
+```bash
+python3 -m training.train_local_policy evaluate --baseline queue_greedy --split val
+```
+
+## TensorBoard
+
+TensorBoard logs are written to `artifacts/dqn_shared/tensorboard` by default.
+
+```bash
+tensorboard --logdir artifacts/dqn_shared/tensorboard
+```
+
+## District LLM
+
+The district LLM stack lives under `district_llm/`. It treats the learned DQN local controller as the low-level executor, derives district-scale SFT labels automatically from DQN rollout windows, and defaults district-model fine-tuning to DQN-derived rows only.
+
+Generate district-LLM data from a learned checkpoint:
+
+```bash
+python3 -m district_llm.generate_dataset \
+  --controller rl_checkpoint \
+  --checkpoint artifacts/dqn_shared/best_validation.pt \
+  --episodes 100 \
+  --decision-interval 10 \
+  --use-checkpoint-env-config \
+  --output data/district_llm_train.jsonl
+```
+
+Generate from fixed or heuristic baselines:
+
+```bash
+python3 -m district_llm.generate_dataset --controller fixed --episodes 50 --decision-interval 10 --output data/district_llm_fixed.jsonl
+python3 -m district_llm.generate_dataset --controller queue_greedy --episodes 50 --decision-interval 10 --output data/district_llm_heuristic.jsonl
+python3 -m district_llm.generate_dataset --teacher-spec fixed --teacher-spec random --episodes 50 --decision-interval 10 --output data/district_llm_multi_teacher.jsonl
+```
+
+Train a first-pass district model with Unsloth/QLoRA:
+
+```bash
+python3 -m training.train_district_llm \
+  --dataset data/district_llm_train.jsonl \
+  --output-dir artifacts/district_llm_qwen \
+  --model-name Qwen/Qwen2.5-7B-Instruct \
+  --load-in-4bit \
+  --lora-rank 16 \
+  --max-seq-length 1024 \
+  --max-steps 1000
+```
+
+Run single-sample inference:
+
+```bash
+python3 -m district_llm.inference \
+  --model artifacts/district_llm_qwen \
+  --city-id city_0006 \
+  --scenario-name accident \
+  --district-id d_00
+```
+
+Run the OpenEnv-compatible district wrapper on top of the current DQN stack:
+
+```bash
+uvicorn openenv_app.app:app --reload
+```
+
+## Algorithm
+
+- Training algorithm: parameter-shared dueling Double DQN.
+- Replay: prioritized replay over per-intersection transitions gathered from full CityFlow worlds.
+- Return target: n-step bootstrap target with target-network updates.
+- Execution: all controllable intersections act simultaneously every RL decision interval.
+- Action space: `0 = hold current phase`, `1 = switch to next green phase`.
+- Safety: `min_green_time` is enforced in the environment and exposed through action masking.
+
+Policy architecture modes:
+
+- `multi_head`: shared trunk with district-type-specific Q heads.
+- `single_head`: one shared Q head for all intersections, with district type removed from the observation.
+- `single_head_with_district_feature`: one shared Q head for all intersections, with district type left in the observation as an explicit feature.
+
+Reward variants:
+
+- `current`: backward-compatible waiting and queue penalty.
+- `normalized_wait_queue`: normalized queue and waiting reduction reward.
+- `wait_queue_throughput`: normalized queue/wait reduction plus throughput bonus and imbalance penalty.
+
+## Smoke Test
+
+To sanity-check one generated scenario with the real CityFlow environment:
+
+```bash
+python3 scripts/smoke_test_env.py --city-id city_0001 --scenario-name normal --policy random
+```
+
+## Project layout
+
+- `agents/`: heuristic local policies and simple baselines.
+- `env/`: CityFlow environment, topology parsing, observation building, and reward logic.
+- `training/`: dataset utilities, replay-based DQN training, evaluation helpers, TensorBoard logging, and CLIs.
+- `data/`: generated synthetic cities, split files, and dataset generation utilities.
+- `scripts/`: utility scripts, including the CityFlow smoke test.
+- `third_party/`: vendored dependencies, including CityFlow source.
+
+## Notes
+
+- The generated dataset is assumed to already exist under `data/generated`.
+- District membership comes from `district_map.json`.
+- District types come from `metadata.json`.
+- Runtime training and evaluation require the `cityflow` Python module to be installed in the active environment.

__init__.py

@@ -0,0 +1 @@
+"""OpenEnv package root for the lean agentic traffic environment."""

agents/README.md

@@ -0,0 +1,20 @@
+# agents
+
+Local traffic-control policies and compatibility shims.
+
+## Main files
+
+- [local_policy.py](/Users/aditya/Developer/traffic-llm/agents/local_policy.py)
+  Active v1 policy interfaces and simple baselines:
+  - `HoldPhasePolicy`
+  - `FixedCyclePolicy`
+  - `QueueGreedyPolicy`
+- [district_controller.py](/Users/aditya/Developer/traffic-llm/agents/district_controller.py)
+  Older district-level prototype logic kept for compatibility.
+- [district_coordinator.py](/Users/aditya/Developer/traffic-llm/agents/district_coordinator.py)
+  Import shim for older code paths.
+
+## Notes
+
+- The learned local-policy network itself lives in [training/models.py](/Users/aditya/Developer/traffic-llm/training/models.py), not here.
+- For active training, use the parameter-shared DQN path in `training/`, not the district-controller prototypes.

agents/__init__.py

@@ -0,0 +1,15 @@
+from agents.local_policy import (
+    BaseLocalPolicy,
+    FixedCyclePolicy,
+    HoldPhasePolicy,
+    QueueGreedyPolicy,
+    SharedHeuristicLocalPolicy,
+)
+
+__all__ = [
+    "BaseLocalPolicy",
+    "FixedCyclePolicy",
+    "HoldPhasePolicy",
+    "QueueGreedyPolicy",
+    "SharedHeuristicLocalPolicy",
+]

agents/district_controller.py

@@ -0,0 +1,187 @@
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from typing import Any, Callable
+
+from agents.message_protocol import DistrictDirective, parse_district_directive
+
+
+class BaseDistrictCoordinator(ABC):
+    @abstractmethod
+    def decide(self, district_summary: dict[str, Any]) -> dict[str, Any]:
+        raise NotImplementedError
+
+
+class RuleBasedDistrictCoordinator(BaseDistrictCoordinator):
+    """
+    Fast, deterministic, and robust.
+    Good first coordinator and good fallback if the LLM output fails.
+    """
+
+    def __init__(
+        self,
+        imbalance_threshold: float = 0.15,
+        border_pressure_threshold: float = 0.65,
+        default_duration: int = 2,
+    ):
+        self.imbalance_threshold = imbalance_threshold
+        self.border_pressure_threshold = border_pressure_threshold
+        self.default_duration = default_duration
+
+    def decide(self, district_summary: dict[str, Any]) -> dict[str, Any]:
+        district_id = district_summary.get("district_id", "unknown")
+        intersection_ids = district_summary.get("intersection_ids", [])
+
+        emergency = district_summary.get("emergency_vehicle", {})
+        if emergency.get("present", False):
+            return (
+                DistrictDirective(
+                    mode="emergency_route",
+                    target_intersections=emergency.get("route", intersection_ids),
+                    duration=2,
+                    rationale=f"Emergency vehicle detected in district {district_id}.",
+                    corridor=emergency.get("corridor"),
+                    district_weight=1.0,
+                )
+                .validate()
+                .to_dict()
+            )
+
+        corridor_loads = district_summary.get("corridor_loads", {})
+        ns = float(corridor_loads.get("ns", corridor_loads.get("north_south", 0.0)))
+        ew = float(corridor_loads.get("ew", corridor_loads.get("east_west", 0.0)))
+
+        border_pressure = district_summary.get("border_pressure", {})
+        border_max = 0.0
+        if isinstance(border_pressure, dict) and border_pressure:
+            border_max = max(float(v) for v in border_pressure.values())
+
+        if ew - ns > self.imbalance_threshold:
+            return (
+                DistrictDirective(
+                    mode="prioritize_ew",
+                    target_intersections=intersection_ids,
+                    duration=self.default_duration,
+                    rationale="East-west corridor is currently more congested than north-south.",
+                    corridor="ew",
+                    district_weight=(
+                        0.7 if border_max < self.border_pressure_threshold else 0.9
+                    ),
+                )
+                .validate()
+                .to_dict()
+            )
+
+        if ns - ew > self.imbalance_threshold:
+            return (
+                DistrictDirective(
+                    mode="prioritize_ns",
+                    target_intersections=intersection_ids,
+                    duration=self.default_duration,
+                    rationale="North-south corridor is currently more congested than east-west.",
+                    corridor="ns",
+                    district_weight=(
+                        0.7 if border_max < self.border_pressure_threshold else 0.9
+                    ),
+                )
+                .validate()
+                .to_dict()
+            )
+
+        if border_max >= self.border_pressure_threshold:
+            return (
+                DistrictDirective(
+                    mode="damp_border_inflow",
+                    target_intersections=intersection_ids,
+                    duration=2,
+                    rationale="Border pressure is high; reduce spill-in and smooth cross-district flow.",
+                    district_weight=0.8,
+                )
+                .validate()
+                .to_dict()
+            )
+
+        return (
+            DistrictDirective(
+                mode="none",
+                target_intersections=[],
+                duration=1,
+                rationale="District is reasonably balanced.",
+                district_weight=0.5,
+            )
+            .validate()
+            .to_dict()
+        )
+
+
+class LLMDistrictCoordinator(BaseDistrictCoordinator):
+    """
+    LLM-backed coordinator.
+
+    `generator_fn` should accept a prompt string and return either:
+      - a JSON string, or
+      - a dict
+
+    Example:
+        coordinator = LLMDistrictCoordinator(generator_fn=my_model_call)
+    """
+
+    def __init__(
+        self,
+        generator_fn: Callable[[str], str | dict[str, Any]],
+        fallback: BaseDistrictCoordinator | None = None,
+        max_prompt_chars: int = 4000,
+    ):
+        self.generator_fn = generator_fn
+        self.fallback = fallback or RuleBasedDistrictCoordinator()
+        self.max_prompt_chars = max_prompt_chars
+
+    def decide(self, district_summary: dict[str, Any]) -> dict[str, Any]:
+        prompt = self.build_prompt(district_summary)
+        try:
+            raw = self.generator_fn(prompt)
+            directive = parse_district_directive(raw).to_dict()
+
+            # If the LLM returns a no-op too often or malformed content,
+            # the parser still makes it safe. We keep that behavior.
+            return directive
+        except Exception:
+            return self.fallback.decide(district_summary)
+
+    def build_prompt(self, district_summary: dict[str, Any]) -> str:
+        summary_text = repr(district_summary)
+        if len(summary_text) > self.max_prompt_chars:
+            summary_text = summary_text[: self.max_prompt_chars] + " ...[truncated]"
+
+        return f"""You are a district-level traffic coordinator.
+
+Your job is to choose a single strategic directive for the next few cycles.
+
+Allowed modes:
+- none
+- prioritize_ns
+- prioritize_ew
+- green_wave
+- emergency_route
+- damp_border_inflow
+
+Return ONLY valid JSON with these fields:
+{{
+  "mode": string,
+  "target_intersections": list[string],
+  "duration": int,
+  "rationale": string,
+  "corridor": string or null,
+  "district_weight": float
+}}
+
+Guidelines:
+- Use emergency_route if an emergency vehicle is present.
+- Use prioritize_ns or prioritize_ew when one corridor is clearly more congested.
+- Use damp_border_inflow when cross-district border pressure is high.
+- Keep duration between 1 and 5.
+- district_weight should be between 0.0 and 1.0.
+
+District summary:
+{summary_text}
+"""

agents/district_coordinator.py

@@ -0,0 +1,11 @@
+from agents.district_controller import (
+    BaseDistrictCoordinator,
+    LLMDistrictCoordinator,
+    RuleBasedDistrictCoordinator,
+)
+
+__all__ = [
+    "BaseDistrictCoordinator",
+    "LLMDistrictCoordinator",
+    "RuleBasedDistrictCoordinator",
+]

agents/heuristic_controller.py

@@ -0,0 +1,91 @@
+from __future__ import annotations
+
+from typing import Any
+
+
+class HeuristicController:
+    """
+    Simple local traffic-light controller.
+
+    Action space:
+        0 -> choose NS green
+        1 -> choose EW green
+
+    Assumes:
+        queue_lengths = [N, S, E, W]
+        waiting_counts = [N, S, E, W]
+    """
+
+    def __init__(
+        self,
+        min_green_steps: int = 5,
+        switch_margin: float = 1.0,
+        district_bonus_scale: float = 3.0,
+        neighbor_pressure_scale: float = 0.25,
+    ):
+        self.min_green_steps = min_green_steps
+        self.switch_margin = switch_margin
+        self.district_bonus_scale = district_bonus_scale
+        self.neighbor_pressure_scale = neighbor_pressure_scale
+
+    def act(self, obs: dict[str, Any]) -> int:
+        queue_lengths = obs.get("queue_lengths", [0, 0, 0, 0])
+        waiting_counts = obs.get("waiting_counts", [0, 0, 0, 0])
+        current_phase = int(obs.get("current_phase", 0))
+        time_since_switch = int(obs.get("time_since_switch", 0))
+        district_mode = obs.get("district_mode", "none")
+        district_weight = float(obs.get("district_weight", 0.5))
+        neighbor_pressure = obs.get("neighbor_pressure", [0.0, 0.0])
+
+        ns_score = (
+            queue_lengths[0]
+            + queue_lengths[1]
+            + 1.5 * (waiting_counts[0] + waiting_counts[1])
+        )
+        ew_score = (
+            queue_lengths[2]
+            + queue_lengths[3]
+            + 1.5 * (waiting_counts[2] + waiting_counts[3])
+        )
+
+        # Optional small neighbor-pressure bias
+        if isinstance(neighbor_pressure, list) and len(neighbor_pressure) >= 2:
+            ns_score += self.neighbor_pressure_scale * float(neighbor_pressure[0])
+            ew_score += self.neighbor_pressure_scale * float(neighbor_pressure[1])
+
+        # District-level strategic bias
+        district_bonus = self.district_bonus_scale * district_weight
+        if district_mode == "prioritize_ns":
+            ns_score += district_bonus
+        elif district_mode == "prioritize_ew":
+            ew_score += district_bonus
+        elif district_mode == "green_wave":
+            corridor = obs.get("district_corridor")
+            if corridor == "ns":
+                ns_score += district_bonus
+            elif corridor == "ew":
+                ew_score += district_bonus
+        elif district_mode == "emergency_route":
+            corridor = obs.get("district_corridor")
+            if corridor in {"north_to_south", "south_to_north", "ns"}:
+                ns_score += district_bonus * 1.5
+            elif corridor in {"west_to_east", "east_to_west", "ew"}:
+                ew_score += district_bonus * 1.5
+
+        desired_phase = 0 if ns_score >= ew_score else 1
+
+        # Avoid thrashing
+        if time_since_switch < self.min_green_steps:
+            return current_phase
+
+        # Only switch if the other phase is meaningfully better
+        current_score = ns_score if current_phase == 0 else ew_score
+        desired_score = ns_score if desired_phase == 0 else ew_score
+
+        if (
+            desired_phase != current_phase
+            and desired_score < current_score + self.switch_margin
+        ):
+            return current_phase
+
+        return desired_phase

agents/local_policy.py

@@ -0,0 +1,110 @@
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+
+import numpy as np
+
+
+class BaseLocalPolicy(ABC):
+    @abstractmethod
+    def act(self, observation_batch: dict[str, np.ndarray]) -> np.ndarray:
+        raise NotImplementedError
+
+
+class HoldPhasePolicy(BaseLocalPolicy):
+    def act(self, observation_batch: dict[str, np.ndarray]) -> np.ndarray:
+        intersection_count = len(observation_batch["intersection_ids"])
+        return np.zeros(intersection_count, dtype=np.int64)
+
+
+class RandomPhasePolicy(BaseLocalPolicy):
+    def __init__(self, seed: int = 7):
+        self.rng = np.random.default_rng(seed)
+
+    def act(self, observation_batch: dict[str, np.ndarray]) -> np.ndarray:
+        action_mask = observation_batch["action_mask"]
+        actions = np.zeros(action_mask.shape[0], dtype=np.int64)
+        for row_index, mask in enumerate(action_mask):
+            valid_actions = np.flatnonzero(mask > 0.0)
+            actions[row_index] = int(self.rng.choice(valid_actions))
+        return actions
+
+
+class FixedCyclePolicy(BaseLocalPolicy):
+    def __init__(self, green_time: int = 20):
+        self.green_time = int(green_time)
+
+    def act(self, observation_batch: dict[str, np.ndarray]) -> np.ndarray:
+        elapsed = observation_batch["phase_elapsed"]
+        action_mask = observation_batch["action_mask"]
+        should_switch = (elapsed >= self.green_time) & (action_mask[:, 1] > 0.0)
+        return should_switch.astype(np.int64)
+
+
+class QueueGreedyPolicy(BaseLocalPolicy):
+    def __init__(self, switch_margin: float = 1.0):
+        self.switch_margin = float(switch_margin)
+
+    def act(self, observation_batch: dict[str, np.ndarray]) -> np.ndarray:
+        counts = observation_batch["incoming_counts"]
+        waiting = observation_batch["incoming_waiting"]
+        lane_mask = observation_batch["lane_mask"]
+        current_phase = observation_batch["current_phase"]
+        action_mask = observation_batch["action_mask"]
+
+        midpoint = counts.shape[1] // 2
+        ns_score = (
+            counts[:, :midpoint].sum(axis=1)
+            + 1.5 * waiting[:, :midpoint].sum(axis=1)
+        )
+        ew_score = (
+            counts[:, midpoint:].sum(axis=1)
+            + 1.5 * waiting[:, midpoint:].sum(axis=1)
+        )
+
+        valid_midpoint = lane_mask[:, :midpoint].sum(axis=1) > 0
+        ns_score = np.where(valid_midpoint, ns_score, 0.0)
+
+        desired_switch = np.where(
+            current_phase == 0,
+            ew_score > ns_score + self.switch_margin,
+            ns_score > ew_score + self.switch_margin,
+        )
+        desired_switch = desired_switch & (action_mask[:, 1] > 0.0)
+        return desired_switch.astype(np.int64)
+
+
+class SharedHeuristicLocalPolicy(QueueGreedyPolicy):
+    def __init__(
+        self,
+        min_green_steps: int = 5,
+        switch_margin: float = 1.0,
+        district_bonus_scale: float = 0.0,
+        neighbor_pressure_scale: float = 0.0,
+    ):
+        self.min_green_steps = int(min_green_steps)
+        del district_bonus_scale, neighbor_pressure_scale
+        super().__init__(switch_margin=switch_margin)
+
+    def act_batch(self, observation_batch):
+        if "intersection_ids" in observation_batch:
+            return self.act(observation_batch)
+
+        actions: dict[str, int] = {}
+        for intersection_id, payload in observation_batch.items():
+            waiting = payload.get("waiting_counts", [0, 0, 0, 0])
+            queues = payload.get("queue_lengths", [0, 0, 0, 0])
+            current_phase = int(payload.get("current_phase", 0))
+            time_since_switch = int(payload.get("time_since_switch", 0))
+
+            ns_score = float(sum(queues[:2]) + 1.5 * sum(waiting[:2]))
+            ew_score = float(sum(queues[2:4]) + 1.5 * sum(waiting[2:4]))
+            desired_phase = 0 if ns_score >= ew_score else 1
+
+            if time_since_switch < self.min_green_steps:
+                actions[intersection_id] = current_phase
+            elif desired_phase != current_phase and abs(ns_score - ew_score) <= self.switch_margin:
+                actions[intersection_id] = current_phase
+            else:
+                actions[intersection_id] = desired_phase
+        return actions

agents/message_protocol.py

@@ -0,0 +1,116 @@
+from __future__ import annotations
+
+import json
+from dataclasses import asdict, dataclass, field
+from typing import Any
+
+
+VALID_MODES = {
+    "none",
+    "prioritize_ns",
+    "prioritize_ew",
+    "green_wave",
+    "emergency_route",
+    "damp_border_inflow",
+}
+
+
+@dataclass
+class NeighborMessage:
+    sender_intersection: str
+    receiver_intersection: str
+    congestion_level: float
+    spillback_risk: bool
+    dominant_direction: str  # "ns", "ew", or "balanced"
+    queue_total: int
+
+    def to_dict(self) -> dict[str, Any]:
+        return asdict(self)
+
+
+@dataclass
+class DistrictDirective:
+    mode: str = "none"
+    target_intersections: list[str] = field(default_factory=list)
+    duration: int = 1
+    rationale: str = ""
+    corridor: str | None = None
+    district_weight: float = 0.5
+
+    def validate(self) -> "DistrictDirective":
+        if self.mode not in VALID_MODES:
+            self.mode = "none"
+
+        if not isinstance(self.target_intersections, list):
+            self.target_intersections = []
+
+        if not isinstance(self.duration, int):
+            self.duration = 1
+        self.duration = max(1, min(self.duration, 10))
+
+        if not isinstance(self.rationale, str):
+            self.rationale = ""
+
+        if self.corridor is not None and self.corridor not in {
+            "ns",
+            "ew",
+            "west_to_east",
+            "east_to_west",
+            "north_to_south",
+            "south_to_north",
+        }:
+            self.corridor = None
+
+        if not isinstance(self.district_weight, (int, float)):
+            self.district_weight = 0.5
+        self.district_weight = float(max(0.0, min(1.0, self.district_weight)))
+
+        return self
+
+    def to_dict(self) -> dict[str, Any]:
+        return asdict(self)
+
+
+def parse_district_directive(payload: str | dict[str, Any]) -> DistrictDirective:
+    """
+    Accept either raw JSON text or a dict and return a validated DistrictDirective.
+    Falls back safely to a no-op directive.
+    """
+    try:
+        if isinstance(payload, str):
+            payload = payload.strip()
+            if not payload:
+                return DistrictDirective().validate()
+
+            # Try direct JSON parse
+            try:
+                data = json.loads(payload)
+            except json.JSONDecodeError:
+                # Try to extract JSON object from surrounding text
+                start = payload.find("{")
+                end = payload.rfind("}")
+                if start == -1 or end == -1 or end <= start:
+                    return DistrictDirective().validate()
+                data = json.loads(payload[start : end + 1])
+        elif isinstance(payload, dict):
+            data = payload
+        else:
+            return DistrictDirective().validate()
+
+        directive = DistrictDirective(
+            mode=data.get("mode", "none"),
+            target_intersections=data.get("target_intersections", []),
+            duration=data.get("duration", 1),
+            rationale=data.get("rationale", ""),
+            corridor=data.get("corridor"),
+            district_weight=data.get("district_weight", 0.5),
+        )
+        return directive.validate()
+    except Exception:
+        return DistrictDirective().validate()
+
+
+def safe_directive_dict(payload: str | dict[str, Any] | None) -> dict[str, Any]:
+    if payload is None:
+        return DistrictDirective().validate().to_dict()
+    return parse_district_directive(payload).to_dict()

artifacts/README.md

@@ -0,0 +1,3 @@
+# Artifacts
+
+For our 3rd iteration of district LLM, 150 is the best.

artifacts/district_llm_adapter_v2/README.md

@@ -0,0 +1,63 @@
+---
+base_model: unsloth/llama-3.1-8b-unsloth-bnb-4bit
+library_name: peft
+model_name: district_llm_adapter_v2
+tags:
+- base_model:adapter:unsloth/llama-3.1-8b-unsloth-bnb-4bit
+- lora
+- sft
+- transformers
+- trl
+- unsloth
+licence: license
+pipeline_tag: text-generation
+---
+
+# Model Card for district_llm_adapter_v2
+
+This model is a fine-tuned version of [unsloth/llama-3.1-8b-unsloth-bnb-4bit](https://huggingface.co/unsloth/llama-3.1-8b-unsloth-bnb-4bit).
+It has been trained using [TRL](https://github.com/huggingface/trl).
+
+## Quick start
+
+```python
+from transformers import pipeline
+
+question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
+generator = pipeline("text-generation", model="None", device="cuda")
+output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
+print(output["generated_text"])
+```
+
+## Training procedure
+
+ 
+
+
+This model was trained with SFT.
+
+### Framework versions
+
+- PEFT 0.18.1
+- TRL: 0.24.0
+- Transformers: 5.2.0
+- Pytorch: 2.10.0
+- Datasets: 4.3.0
+- Tokenizers: 0.22.2
+
+## Citations
+
+
+
+Cite TRL as:
+    
+```bibtex
+@misc{vonwerra2022trl,
+	title        = {{TRL: Transformer Reinforcement Learning}},
+	author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec},
+	year         = 2020,
+	journal      = {GitHub repository},
+	publisher    = {GitHub},
+	howpublished = {\url{https://github.com/huggingface/trl}}
+}
+```

artifacts/district_llm_adapter_v2/adapter_config.json

@@ -0,0 +1,50 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "LlamaForCausalLM",
+    "parent_library": "transformers.models.llama.modeling_llama",
+    "unsloth_fixed": true
+  },
+  "base_model_name_or_path": "unsloth/llama-3.1-8b-unsloth-bnb-4bit",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "up_proj",
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

artifacts/district_llm_adapter_v2/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18e9b835a17a3b1429550ea5e4afaa9b8ecc9301da476f35ece0679c9fb0203a
+size 167832240

artifacts/district_llm_adapter_v2/checkpoint-100/README.md

@@ -0,0 +1,210 @@
+---
+base_model: unsloth/llama-3.1-8b-unsloth-bnb-4bit
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:unsloth/llama-3.1-8b-unsloth-bnb-4bit
+- lora
+- sft
+- transformers
+- trl
+- unsloth
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1

artifacts/district_llm_adapter_v2/checkpoint-100/adapter_config.json

@@ -0,0 +1,50 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "LlamaForCausalLM",
+    "parent_library": "transformers.models.llama.modeling_llama",
+    "unsloth_fixed": true
+  },
+  "base_model_name_or_path": "unsloth/llama-3.1-8b-unsloth-bnb-4bit",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "up_proj",
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

artifacts/district_llm_adapter_v2/checkpoint-100/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccc17edb05ab2ceffb808c495859a48e7eb78507264a0404bc0aa2cda8eab3f3
+size 167832240

artifacts/district_llm_adapter_v2/checkpoint-100/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0e7cf928e3b2955b39d6a9e5b96926f624d9f53efd6b1703fdc8658bd792b5a
+size 85728229

artifacts/district_llm_adapter_v2/checkpoint-100/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c800b778fa7e115e4c34de8529902de8b61c9a1b4bab3eb8295d06dafff030e
+size 14645

artifacts/district_llm_adapter_v2/checkpoint-100/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f48b2993192172e90d469975ad49cfb293d165a9bc6f65da704bfab6cd3cab3
+size 1465

artifacts/district_llm_adapter_v2/checkpoint-100/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b9e4e7fb171f92fd137b777cc2714bf87d11576700a1dcd7a399e7bbe39537b
+size 17209920

artifacts/district_llm_adapter_v2/checkpoint-100/tokenizer_config.json

@@ -0,0 +1,18 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|end_of_text|>",
+  "from_slow": true,
+  "is_local": false,
+  "legacy": false,
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 131072,
+  "pad_token": "<|finetune_right_pad_id|>",
+  "padding_side": "right",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null
+}

artifacts/district_llm_adapter_v2/checkpoint-100/trainer_state.json

@@ -0,0 +1,174 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.5333333333333333,
+  "eval_steps": 50,
+  "global_step": 100,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.02666666666666667,
+      "grad_norm": 0.824774444103241,
+      "learning_rate": 4e-05,
+      "loss": 1.4378397941589356,
+      "step": 5
+    },
+    {
+      "epoch": 0.05333333333333334,
+      "grad_norm": 1.7167983055114746,
+      "learning_rate": 9e-05,
+      "loss": 1.087998867034912,
+      "step": 10
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.7535544037818909,
+      "learning_rate": 0.00014,
+      "loss": 0.5613192558288574,
+      "step": 15
+    },
+    {
+      "epoch": 0.10666666666666667,
+      "grad_norm": 0.6983357071876526,
+      "learning_rate": 0.00019,
+      "loss": 0.28872098922729494,
+      "step": 20
+    },
+    {
+      "epoch": 0.13333333333333333,
+      "grad_norm": 0.8250440955162048,
+      "learning_rate": 0.00019989930665413147,
+      "loss": 0.22604494094848632,
+      "step": 25
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.26267266273498535,
+      "learning_rate": 0.00019949058745487522,
+      "loss": 0.20742559432983398,
+      "step": 30
+    },
+    {
+      "epoch": 0.18666666666666668,
+      "grad_norm": 0.27337217330932617,
+      "learning_rate": 0.00019876883405951377,
+      "loss": 0.1870889902114868,
+      "step": 35
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 0.10534920543432236,
+      "learning_rate": 0.00019773631737125192,
+      "loss": 0.18097405433654784,
+      "step": 40
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.13477347791194916,
+      "learning_rate": 0.00019639628606958533,
+      "loss": 0.17958487272262574,
+      "step": 45
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 0.1370360553264618,
+      "learning_rate": 0.0001947529563887529,
+      "loss": 0.16803257465362548,
+      "step": 50
+    },
+    {
+      "epoch": 0.29333333333333333,
+      "grad_norm": 0.1513640433549881,
+      "learning_rate": 0.0001928114988519039,
+      "loss": 0.16801449060440063,
+      "step": 55
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.11957086622714996,
+      "learning_rate": 0.00019057802200271942,
+      "loss": 0.16269906759262084,
+      "step": 60
+    },
+    {
+      "epoch": 0.3466666666666667,
+      "grad_norm": 0.11077677458524704,
+      "learning_rate": 0.0001880595531856738,
+      "loss": 0.16706794500350952,
+      "step": 65
+    },
+    {
+      "epoch": 0.37333333333333335,
+      "grad_norm": 0.10614161193370819,
+      "learning_rate": 0.00018526401643540922,
+      "loss": 0.16436902284622193,
+      "step": 70
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.13366934657096863,
+      "learning_rate": 0.00018220020754479102,
+      "loss": 0.15710221529006957,
+      "step": 75
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 0.1160494014620781,
+      "learning_rate": 0.00017887776639008914,
+      "loss": 0.16149884462356567,
+      "step": 80
+    },
+    {
+      "epoch": 0.4533333333333333,
+      "grad_norm": 0.09502895176410675,
+      "learning_rate": 0.00017530714660036112,
+      "loss": 0.16155229806900023,
+      "step": 85
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.11183392256498337,
+      "learning_rate": 0.00017149958266646754,
+      "loss": 0.15782997608184815,
+      "step": 90
+    },
+    {
+      "epoch": 0.5066666666666667,
+      "grad_norm": 0.11126931011676788,
+      "learning_rate": 0.00016746705459320745,
+      "loss": 0.1604154586791992,
+      "step": 95
+    },
+    {
+      "epoch": 0.5333333333333333,
+      "grad_norm": 0.0941988080739975,
+      "learning_rate": 0.00016322225020579099,
+      "loss": 0.15799541473388673,
+      "step": 100
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 300,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 50,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.3346529297301504e+16,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

artifacts/district_llm_adapter_v2/checkpoint-100/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99acc664411451a603c8908d063cb9b0bf85f277436eb270d65909aea5df4002
+size 5777

artifacts/district_llm_adapter_v2/checkpoint-150/README.md

@@ -0,0 +1,210 @@
+---
+base_model: unsloth/llama-3.1-8b-unsloth-bnb-4bit
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:unsloth/llama-3.1-8b-unsloth-bnb-4bit
+- lora
+- sft
+- transformers
+- trl
+- unsloth
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1

artifacts/district_llm_adapter_v2/checkpoint-150/adapter_config.json

@@ -0,0 +1,50 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "LlamaForCausalLM",
+    "parent_library": "transformers.models.llama.modeling_llama",
+    "unsloth_fixed": true
+  },
+  "base_model_name_or_path": "unsloth/llama-3.1-8b-unsloth-bnb-4bit",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "up_proj",
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

artifacts/district_llm_adapter_v2/checkpoint-150/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b43a432bd838d3078559afa204238c64fb87d6d49aa0f53747daa3495f20a201
+size 167832240

artifacts/district_llm_adapter_v2/checkpoint-150/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8b98930b82c827773787056dc5ba8fcecee40c8b3331dd32053627319e71fee
+size 85728229

artifacts/district_llm_adapter_v2/checkpoint-150/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c800b778fa7e115e4c34de8529902de8b61c9a1b4bab3eb8295d06dafff030e
+size 14645

artifacts/district_llm_adapter_v2/checkpoint-150/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f9a3661913848bed826a840dc7f1ea0f7872a368500bea34bbe970fd6d04f2e
+size 1465

artifacts/district_llm_adapter_v2/checkpoint-150/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b9e4e7fb171f92fd137b777cc2714bf87d11576700a1dcd7a399e7bbe39537b
+size 17209920

artifacts/district_llm_adapter_v2/checkpoint-150/tokenizer_config.json

@@ -0,0 +1,18 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|end_of_text|>",
+  "from_slow": true,
+  "is_local": false,
+  "legacy": false,
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 131072,
+  "pad_token": "<|finetune_right_pad_id|>",
+  "padding_side": "right",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null
+}

artifacts/district_llm_adapter_v2/checkpoint-150/trainer_state.json

@@ -0,0 +1,244 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.8,
+  "eval_steps": 50,
+  "global_step": 150,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.02666666666666667,
+      "grad_norm": 0.824774444103241,
+      "learning_rate": 4e-05,
+      "loss": 1.4378397941589356,
+      "step": 5
+    },
+    {
+      "epoch": 0.05333333333333334,
+      "grad_norm": 1.7167983055114746,
+      "learning_rate": 9e-05,
+      "loss": 1.087998867034912,
+      "step": 10
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.7535544037818909,
+      "learning_rate": 0.00014,
+      "loss": 0.5613192558288574,
+      "step": 15
+    },
+    {
+      "epoch": 0.10666666666666667,
+      "grad_norm": 0.6983357071876526,
+      "learning_rate": 0.00019,
+      "loss": 0.28872098922729494,
+      "step": 20
+    },
+    {
+      "epoch": 0.13333333333333333,
+      "grad_norm": 0.8250440955162048,
+      "learning_rate": 0.00019989930665413147,
+      "loss": 0.22604494094848632,
+      "step": 25
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.26267266273498535,
+      "learning_rate": 0.00019949058745487522,
+      "loss": 0.20742559432983398,
+      "step": 30
+    },
+    {
+      "epoch": 0.18666666666666668,
+      "grad_norm": 0.27337217330932617,
+      "learning_rate": 0.00019876883405951377,
+      "loss": 0.1870889902114868,
+      "step": 35
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 0.10534920543432236,
+      "learning_rate": 0.00019773631737125192,
+      "loss": 0.18097405433654784,
+      "step": 40
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.13477347791194916,
+      "learning_rate": 0.00019639628606958533,
+      "loss": 0.17958487272262574,
+      "step": 45
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 0.1370360553264618,
+      "learning_rate": 0.0001947529563887529,
+      "loss": 0.16803257465362548,
+      "step": 50
+    },
+    {
+      "epoch": 0.29333333333333333,
+      "grad_norm": 0.1513640433549881,
+      "learning_rate": 0.0001928114988519039,
+      "loss": 0.16801449060440063,
+      "step": 55
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.11957086622714996,
+      "learning_rate": 0.00019057802200271942,
+      "loss": 0.16269906759262084,
+      "step": 60
+    },
+    {
+      "epoch": 0.3466666666666667,
+      "grad_norm": 0.11077677458524704,
+      "learning_rate": 0.0001880595531856738,
+      "loss": 0.16706794500350952,
+      "step": 65
+    },
+    {
+      "epoch": 0.37333333333333335,
+      "grad_norm": 0.10614161193370819,
+      "learning_rate": 0.00018526401643540922,
+      "loss": 0.16436902284622193,
+      "step": 70
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.13366934657096863,
+      "learning_rate": 0.00018220020754479102,
+      "loss": 0.15710221529006957,
+      "step": 75
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 0.1160494014620781,
+      "learning_rate": 0.00017887776639008914,
+      "loss": 0.16149884462356567,
+      "step": 80
+    },
+    {
+      "epoch": 0.4533333333333333,
+      "grad_norm": 0.09502895176410675,
+      "learning_rate": 0.00017530714660036112,
+      "loss": 0.16155229806900023,
+      "step": 85
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.11183392256498337,
+      "learning_rate": 0.00017149958266646754,
+      "loss": 0.15782997608184815,
+      "step": 90
+    },
+    {
+      "epoch": 0.5066666666666667,
+      "grad_norm": 0.11126931011676788,
+      "learning_rate": 0.00016746705459320745,
+      "loss": 0.1604154586791992,
+      "step": 95
+    },
+    {
+      "epoch": 0.5333333333333333,
+      "grad_norm": 0.0941988080739975,
+      "learning_rate": 0.00016322225020579099,
+      "loss": 0.15799541473388673,
+      "step": 100
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.1058623343706131,
+      "learning_rate": 0.00015877852522924732,
+      "loss": 0.15690511465072632,
+      "step": 105
+    },
+    {
+      "epoch": 0.5866666666666667,
+      "grad_norm": 0.11555207520723343,
+      "learning_rate": 0.00015414986126637258,
+      "loss": 0.1615644574165344,
+      "step": 110
+    },
+    {
+      "epoch": 0.6133333333333333,
+      "grad_norm": 0.2579882740974426,
+      "learning_rate": 0.0001493508218064347,
+      "loss": 0.16159408092498778,
+      "step": 115
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11453735083341599,
+      "learning_rate": 0.00014439650640304822,
+      "loss": 0.1569008708000183,
+      "step": 120
+    },
+    {
+      "epoch": 0.6666666666666666,
+      "grad_norm": 0.12122868746519089,
+      "learning_rate": 0.00013930250316539238,
+      "loss": 0.15291671752929686,
+      "step": 125
+    },
+    {
+      "epoch": 0.6933333333333334,
+      "grad_norm": 0.12676002085208893,
+      "learning_rate": 0.0001340848397122525,
+      "loss": 0.14967869520187377,
+      "step": 130
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.14859774708747864,
+      "learning_rate": 0.00012875993274320173,
+      "loss": 0.1521363615989685,
+      "step": 135
+    },
+    {
+      "epoch": 0.7466666666666667,
+      "grad_norm": 0.1960573047399521,
+      "learning_rate": 0.00012334453638559057,
+      "loss": 0.14652782678604126,
+      "step": 140
+    },
+    {
+      "epoch": 0.7733333333333333,
+      "grad_norm": 0.13388624787330627,
+      "learning_rate": 0.00011785568947986367,
+      "loss": 0.1499272108078003,
+      "step": 145
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.16420726478099823,
+      "learning_rate": 0.0001123106619690643,
+      "loss": 0.14629043340682985,
+      "step": 150
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 300,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 50,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.921437712999219e+16,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

artifacts/district_llm_adapter_v2/checkpoint-150/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99acc664411451a603c8908d063cb9b0bf85f277436eb270d65909aea5df4002
+size 5777

artifacts/district_llm_adapter_v2/checkpoint-200/README.md

@@ -0,0 +1,210 @@
+---
+base_model: unsloth/llama-3.1-8b-unsloth-bnb-4bit
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:unsloth/llama-3.1-8b-unsloth-bnb-4bit
+- lora
+- sft
+- transformers
+- trl
+- unsloth
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1

artifacts/district_llm_adapter_v2/checkpoint-200/adapter_config.json

@@ -0,0 +1,50 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "LlamaForCausalLM",
+    "parent_library": "transformers.models.llama.modeling_llama",
+    "unsloth_fixed": true
+  },
+  "base_model_name_or_path": "unsloth/llama-3.1-8b-unsloth-bnb-4bit",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "up_proj",
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

artifacts/district_llm_adapter_v2/checkpoint-200/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb70b0b07fb3077ba18082208727936bd4e80e731772e0483ce750854c14a75f
+size 167832240

artifacts/district_llm_adapter_v2/checkpoint-200/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b10686233c23619155b5c07584d46c5a14e2afb65b7140a3782c4ee832a667df
+size 85728229

artifacts/district_llm_adapter_v2/checkpoint-200/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:181c5f0270cf39930062ddfa3767a2481d0c360f120b11f8e25dbf533a1cdaba
+size 14645

artifacts/district_llm_adapter_v2/checkpoint-200/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:379326d22f10fb05f8f610d4844919bdb60346c7b0ec96a2613fb34ff5180597
+size 1465

artifacts/district_llm_adapter_v2/checkpoint-200/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b9e4e7fb171f92fd137b777cc2714bf87d11576700a1dcd7a399e7bbe39537b
+size 17209920

artifacts/district_llm_adapter_v2/checkpoint-200/tokenizer_config.json

@@ -0,0 +1,18 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|end_of_text|>",
+  "from_slow": true,
+  "is_local": false,
+  "legacy": false,
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 131072,
+  "pad_token": "<|finetune_right_pad_id|>",
+  "padding_side": "right",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null
+}

artifacts/district_llm_adapter_v2/checkpoint-200/trainer_state.json

@@ -0,0 +1,314 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.064,
+  "eval_steps": 50,
+  "global_step": 200,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.02666666666666667,
+      "grad_norm": 0.824774444103241,
+      "learning_rate": 4e-05,
+      "loss": 1.4378397941589356,
+      "step": 5
+    },
+    {
+      "epoch": 0.05333333333333334,
+      "grad_norm": 1.7167983055114746,
+      "learning_rate": 9e-05,
+      "loss": 1.087998867034912,
+      "step": 10
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 0.7535544037818909,
+      "learning_rate": 0.00014,
+      "loss": 0.5613192558288574,
+      "step": 15
+    },
+    {
+      "epoch": 0.10666666666666667,
+      "grad_norm": 0.6983357071876526,
+      "learning_rate": 0.00019,
+      "loss": 0.28872098922729494,
+      "step": 20
+    },
+    {
+      "epoch": 0.13333333333333333,
+      "grad_norm": 0.8250440955162048,
+      "learning_rate": 0.00019989930665413147,
+      "loss": 0.22604494094848632,
+      "step": 25
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 0.26267266273498535,
+      "learning_rate": 0.00019949058745487522,
+      "loss": 0.20742559432983398,
+      "step": 30
+    },
+    {
+      "epoch": 0.18666666666666668,
+      "grad_norm": 0.27337217330932617,
+      "learning_rate": 0.00019876883405951377,
+      "loss": 0.1870889902114868,
+      "step": 35
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 0.10534920543432236,
+      "learning_rate": 0.00019773631737125192,
+      "loss": 0.18097405433654784,
+      "step": 40
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.13477347791194916,
+      "learning_rate": 0.00019639628606958533,
+      "loss": 0.17958487272262574,
+      "step": 45
+    },
+    {
+      "epoch": 0.26666666666666666,
+      "grad_norm": 0.1370360553264618,
+      "learning_rate": 0.0001947529563887529,
+      "loss": 0.16803257465362548,
+      "step": 50
+    },
+    {
+      "epoch": 0.29333333333333333,
+      "grad_norm": 0.1513640433549881,
+      "learning_rate": 0.0001928114988519039,
+      "loss": 0.16801449060440063,
+      "step": 55
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.11957086622714996,
+      "learning_rate": 0.00019057802200271942,
+      "loss": 0.16269906759262084,
+      "step": 60
+    },
+    {
+      "epoch": 0.3466666666666667,
+      "grad_norm": 0.11077677458524704,
+      "learning_rate": 0.0001880595531856738,
+      "loss": 0.16706794500350952,
+      "step": 65
+    },
+    {
+      "epoch": 0.37333333333333335,
+      "grad_norm": 0.10614161193370819,
+      "learning_rate": 0.00018526401643540922,
+      "loss": 0.16436902284622193,
+      "step": 70
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 0.13366934657096863,
+      "learning_rate": 0.00018220020754479102,
+      "loss": 0.15710221529006957,
+      "step": 75
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 0.1160494014620781,
+      "learning_rate": 0.00017887776639008914,
+      "loss": 0.16149884462356567,
+      "step": 80
+    },
+    {
+      "epoch": 0.4533333333333333,
+      "grad_norm": 0.09502895176410675,
+      "learning_rate": 0.00017530714660036112,
+      "loss": 0.16155229806900023,
+      "step": 85
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.11183392256498337,
+      "learning_rate": 0.00017149958266646754,
+      "loss": 0.15782997608184815,
+      "step": 90
+    },
+    {
+      "epoch": 0.5066666666666667,
+      "grad_norm": 0.11126931011676788,
+      "learning_rate": 0.00016746705459320745,
+      "loss": 0.1604154586791992,
+      "step": 95
+    },
+    {
+      "epoch": 0.5333333333333333,
+      "grad_norm": 0.0941988080739975,
+      "learning_rate": 0.00016322225020579099,
+      "loss": 0.15799541473388673,
+      "step": 100
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 0.1058623343706131,
+      "learning_rate": 0.00015877852522924732,
+      "loss": 0.15690511465072632,
+      "step": 105
+    },
+    {
+      "epoch": 0.5866666666666667,
+      "grad_norm": 0.11555207520723343,
+      "learning_rate": 0.00015414986126637258,
+      "loss": 0.1615644574165344,
+      "step": 110
+    },
+    {
+      "epoch": 0.6133333333333333,
+      "grad_norm": 0.2579882740974426,
+      "learning_rate": 0.0001493508218064347,
+      "loss": 0.16159408092498778,
+      "step": 115
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11453735083341599,
+      "learning_rate": 0.00014439650640304822,
+      "loss": 0.1569008708000183,
+      "step": 120
+    },
+    {
+      "epoch": 0.6666666666666666,
+      "grad_norm": 0.12122868746519089,
+      "learning_rate": 0.00013930250316539238,
+      "loss": 0.15291671752929686,
+      "step": 125
+    },
+    {
+      "epoch": 0.6933333333333334,
+      "grad_norm": 0.12676002085208893,
+      "learning_rate": 0.0001340848397122525,
+      "loss": 0.14967869520187377,
+      "step": 130
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 0.14859774708747864,
+      "learning_rate": 0.00012875993274320173,
+      "loss": 0.1521363615989685,
+      "step": 135
+    },
+    {
+      "epoch": 0.7466666666666667,
+      "grad_norm": 0.1960573047399521,
+      "learning_rate": 0.00012334453638559057,
+      "loss": 0.14652782678604126,
+      "step": 140
+    },
+    {
+      "epoch": 0.7733333333333333,
+      "grad_norm": 0.13388624787330627,
+      "learning_rate": 0.00011785568947986367,
+      "loss": 0.1499272108078003,
+      "step": 145
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.16420726478099823,
+      "learning_rate": 0.0001123106619690643,
+      "loss": 0.14629043340682985,
+      "step": 150
+    },
+    {
+      "epoch": 0.8266666666666667,
+      "grad_norm": 0.14654554426670074,
+      "learning_rate": 0.00010672690056120399,
+      "loss": 0.14661697149276734,
+      "step": 155
+    },
+    {
+      "epoch": 0.8533333333333334,
+      "grad_norm": 0.14784203469753265,
+      "learning_rate": 0.0001011219738354646,
+      "loss": 0.13817673921585083,
+      "step": 160
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 0.13851715624332428,
+      "learning_rate": 9.551351696494854e-05,
+      "loss": 0.14781265258789061,
+      "step": 165
+    },
+    {
+      "epoch": 0.9066666666666666,
+      "grad_norm": 0.17259690165519714,
+      "learning_rate": 8.991917622989956e-05,
+      "loss": 0.14397273063659669,
+      "step": 170
+    },
+    {
+      "epoch": 0.9333333333333333,
+      "grad_norm": 0.287122517824173,
+      "learning_rate": 8.435655349597689e-05,
+      "loss": 0.14055378437042237,
+      "step": 175
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.15599672496318817,
+      "learning_rate": 7.884315083227373e-05,
+      "loss": 0.14050980806350707,
+      "step": 180
+    },
+    {
+      "epoch": 0.9866666666666667,
+      "grad_norm": 0.17161861062049866,
+      "learning_rate": 7.339631544333249e-05,
+      "loss": 0.13583474159240722,
+      "step": 185
+    },
+    {
+      "epoch": 1.0106666666666666,
+      "grad_norm": 0.17195405066013336,
+      "learning_rate": 6.803318508842187e-05,
+      "loss": 0.14056421518325807,
+      "step": 190
+    },
+    {
+      "epoch": 1.0373333333333334,
+      "grad_norm": 0.14843153953552246,
+      "learning_rate": 6.277063415980549e-05,
+      "loss": 0.14321362972259521,
+      "step": 195
+    },
+    {
+      "epoch": 1.064,
+      "grad_norm": 0.17835235595703125,
+      "learning_rate": 5.762522058966113e-05,
+      "loss": 0.13067924976348877,
+      "step": 200
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 300,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 50,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.496001098695475e+16,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

artifacts/district_llm_adapter_v2/checkpoint-200/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99acc664411451a603c8908d063cb9b0bf85f277436eb270d65909aea5df4002
+size 5777

artifacts/district_llm_adapter_v2/checkpoint-250/README.md

@@ -0,0 +1,210 @@
+---
+base_model: unsloth/llama-3.1-8b-unsloth-bnb-4bit
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:unsloth/llama-3.1-8b-unsloth-bnb-4bit
+- lora
+- sft
+- transformers
+- trl
+- unsloth
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.18.1

artifacts/district_llm_adapter_v2/checkpoint-250/adapter_config.json

@@ -0,0 +1,50 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "LlamaForCausalLM",
+    "parent_library": "transformers.models.llama.modeling_llama",
+    "unsloth_fixed": true
+  },
+  "base_model_name_or_path": "unsloth/llama-3.1-8b-unsloth-bnb-4bit",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "up_proj",
+    "q_proj",
+    "k_proj",
+    "v_proj",
+    "o_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}