Instructions to use Yuchiwang02/Llama-3.2-1B-DelaySentinel with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Yuchiwang02/Llama-3.2-1B-DelaySentinel with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Yuchiwang02/Llama-3.2-1B-DelaySentinel")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Yuchiwang02/Llama-3.2-1B-DelaySentinel")
model = AutoModelForCausalLM.from_pretrained("Yuchiwang02/Llama-3.2-1B-DelaySentinel", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Yuchiwang02/Llama-3.2-1B-DelaySentinel with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Yuchiwang02/Llama-3.2-1B-DelaySentinel"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Yuchiwang02/Llama-3.2-1B-DelaySentinel",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Yuchiwang02/Llama-3.2-1B-DelaySentinel

SGLang

How to use Yuchiwang02/Llama-3.2-1B-DelaySentinel with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Yuchiwang02/Llama-3.2-1B-DelaySentinel" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Yuchiwang02/Llama-3.2-1B-DelaySentinel",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Yuchiwang02/Llama-3.2-1B-DelaySentinel" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Yuchiwang02/Llama-3.2-1B-DelaySentinel",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Yuchiwang02/Llama-3.2-1B-DelaySentinel with Docker Model Runner:
```
docker model run hf.co/Yuchiwang02/Llama-3.2-1B-DelaySentinel
```

Llama-3.2-1B-DelaySentinel

Building and evaluating AI on logistics data. Built with Llama.

This checkpoint is a full-parameter fine-tune of meta-llama/Llama-3.2-1B-Instruct, published with the data split, evaluation results, and behavioral tests used to inspect it. The project connects a working training-and-inference pipeline with a retrospective audit of what its score demonstrates.

The checkpoint and a depth-2 decision tree both score 100% on the same historical 200-row split. A rule using two supplied fields reproduces every label in the 1,000-row source table. Prompt rewrites reveal additional response failures, including sensitivity to irrelevant text. These findings make the checkpoint useful for studying evaluation and label leakage.

GitHub project · Case study · Full evaluation · Frozen dataset · Reproduce

The figure summarizes a historical, partly exposed split: its first 20 rows were used for training-time eval loss. It is a baseline comparison for this artifact, not an independent estimate of prospective forecasting performance.

Model details

item	recorded setup
base checkpoint	`meta-llama/Llama-3.2-1B-Instruct`
method	full-parameter supervised fine-tuning
trainable parameters	1,235,814,400
training data	800 records from a 1,000-row Kaggle logistics table
historical evaluation file	200 records; the first 20 were used for eval loss during training
input	system instruction plus 15 `Column: value` lines
output	`Logistics_Delay: 0` or `Logistics_Delay: 1`
weights trained	September 2025
retrospective audit	September 2026
weights license	Llama 3.2 Community License and Acceptable Use Policy

The frozen split was created with an unseeded shuffle. It can be checked against recorded hashes, but the original shuffle cannot be regenerated. Because the first 20 evaluation rows were exposed during training-time eval loss, the historical file is not a clean independent test set. See split provenance.

Evaluation summary

predictor, same historical 200 rows	accuracy	F1	accuracy interval, 95% Wilson
fine-tuned checkpoint, greedy decoding	1.000	1.000	[0.981, 1.000]
depth-2 decision tree	1.000	1.000	[0.981, 1.000]
all-positive baseline	0.580	0.734	[0.511, 0.646]

Source: results/eval.json, model.test_metrics_greedy and baselines_split_v0. The target satisfies Shipment_Status == "Delayed" OR Traffic_Status == "Heavy" on all 1,000 source rows. The table measures agreement with that dataset label; it does not establish future delay prediction or an advantage over classical models.

A behavioral test appends "Note: the depot supervisor is Mr. Delayed" to the 84 originally negative prompts without changing their logistics fields. Every output changes to 1. The full audit records 104 related probe sets; they reuse source prompts and are not independent experiments. The tests describe observed response patterns, while the internal mechanism remains unknown. See the complete probe tables.

The separate Olist reference study uses classical models on real orders. Its results are not an evaluation of this Llama checkpoint.

Intended use and scope

Use this checkpoint to reproduce a label-leakage case study, inspect behavior under controlled input rewrites, or teach baseline comparison and evaluation design. The repository provides recorded outputs and executable checks for these purposes.

It has not been validated for operational delay forecasting, risk scoring, or dispatch decisions. It provides hard labels rather than calibrated risk estimates. Tested invalid or incomplete inputs can still receive a label instead of an abstention. Examples and probes are specific to this checkpoint and prompt format; they do not establish general conclusions about LLMs on tabular data.

How to use

The checkpoint was trained on the 15-column schema below. The example reproduces an original-format response; it is not an interface validated for real-world risk prediction. Decode greedily, pin the template date, and parse with an anchored regex; treat a non-match as unparsable, never as 0.

import re
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

REPO = "Yuchiwang02/Llama-3.2-1B-DelaySentinel"
tok = AutoTokenizer.from_pretrained(REPO)
model = AutoModelForCausalLM.from_pretrained(REPO, torch_dtype=torch.bfloat16, device_map="auto")

SYSTEM = ("Assume you are a supply chain analyst. Based on the following information, "
          "output the result for Logistics_Delay, where 1 represents a delay and 0 represents no delay.")
# First row of data/test.jsonl (gold label 0). Same 15 columns, same order, as in training.
USER = """Timestamp: 2024-07-18 12:19:20
Asset_ID: Truck_10
Latitude: 20.2969
Longitude: 124.1885
Inventory_Level: 298
Shipment_Status: In Transit
Temperature: 18.6
Humidity: 74.4
Traffic_Status: Detour
Waiting_Time: 41
User_Transaction_Amount: 248
User_Purchase_Frequency: 1
Logistics_Delay_Reason: None
Asset_Utilization: 86.3
Demand_Forecast: 283"""

messages = [{"role": "system", "content": SYSTEM}, {"role": "user", "content": USER}]
input_ids = tok.apply_chat_template(
    messages, add_generation_prompt=True, return_tensors="pt", date_string="04 Sep 2025"
).to(model.device)
out = model.generate(
    input_ids,
    do_sample=False,
    max_new_tokens=8,
    pad_token_id=tok.convert_tokens_to_ids("<|finetune_right_pad_id|>"),  # never add a pad token / resize embeddings
)
text = tok.decode(out[0, input_ids.shape[1]:], skip_special_tokens=True).strip()
m = re.match(r"^\s*Logistics_Delay:\s*([01])\b", text)
label = int(m.group(1)) if m else None  # None = unparsable
print(text, label)                      # -> Logistics_Delay: 0  0

Or, with the package: from delaysentinel.model import Scorer and Scorer(REPO).predict([USER]).

Training and data

The source is Kaggle's Smart Logistics Supply Chain Dataset by ziya07, listed as CC0 when recorded by the project. It contains 1,000 timestamped records, 15 supplied fields, and the target. A deterministic target rule and geographically dispersed coordinates are consistent with a constructed table, but the repository does not contain a generator or independently verified collection provenance. Promised and actual delivery dates are absent.

The original run used 30 epochs, 1,500 steps, bf16, batch size 4 with gradient accumulation 4, learning rate 2e-5, cosine scheduling, and assistant-only loss. Training configuration seed 42 does not recover the earlier unseeded data shuffle. The logged training loss rounds to 0.0000 at step 50; this is a rounded log value, not an assertion of mathematically zero loss.

Training used an adaptation of acon96/home-llm with three local changes and an attribution header. Recovered settings and preserved artifacts are described in run provenance and the technical report. The script is available in the GitHub source.

Limitations retained for reproducibility

The source label is reconstructible from supplied inputs and lacks a documented prospective delivery outcome. The historical split was partly exposed during training-time evaluation.
Behavioral probes reveal sensitivity to wording, negation, and irrelevant values; the internal mechanism and the full set of possible triggers are not established.
The historical collator masked genuine system/user end-of-turn tokens along with padding. This training artifact is documented; no retrained comparison establishes its causal impact.
No checkpoint before step 1,300 survives, preventing analysis of when the final behavior arose.
No separate validation period or calibration study establishes a business decision threshold for this checkpoint. Updated inference examples do not change the historical training process.
Recorded bf16 margins can vary in their final digits across batch composition and environments.

The evaluation report contains the complete methods, tables, probe conditions, and evidence boundaries. The reproduction guide separates CPU checks, full model execution, and the independent Olist study.

Attribution and licenses

Built with Llama. The weights are subject to the Llama 3.2 Community License and Acceptable Use Policy, with attribution in NOTICE. The full model name is Llama-3.2-1B-DelaySentinel; the original Hub name redirects to this repository.

The training script derives from acon96/home-llm under MIT, with Stanford Alpaca portions under Apache-2.0. The preserved copy has three local changes relative to the identified upstream revision; see third-party attribution and the Apache license.

Project code and documentation are distributed under MIT. The Olist aggregate result is distributed under CC BY-NC-SA 4.0. Implementation and documentation use AI assistance; linked source records and the attribution file describe the project materials and reused work.

Citation

@misc{wang2026delaysentinel,
  title  = {Llama-3.2-1B-DelaySentinel: Building and evaluating AI on logistics data},
  author = {Wang, Yuchi},
  year   = {2026},
  note   = {Weights trained September 2025; retrospective evaluation and behavioral audit added September 2026},
  url    = {https://huggingface.co/Yuchiwang02/Llama-3.2-1B-DelaySentinel}
}

中文案例 · Related project: BizHallu

Downloads last month: 535

Safetensors

Model size

1B params

Tensor type

BF16

Model tree for Yuchiwang02/Llama-3.2-1B-DelaySentinel

Base model

meta-llama/Llama-3.2-1B-Instruct

Finetuned

(1804)

this model

Yuchiwang02
/

Llama-3.2-1B-DelaySentinel