Instructions to use lldois/v102_v74_pcgrad_normalized_task_delta with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries
PEFT
How to use lldois/v102_v74_pcgrad_normalized_task_delta with PEFT:
```
Task type is invalid.
```

How to use lldois/v102_v74_pcgrad_normalized_task_delta with Transformers:

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

pipe = pipeline("text-generation", model="lldois/v102_v74_pcgrad_normalized_task_delta")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("lldois/v102_v74_pcgrad_normalized_task_delta")
model = AutoModelForCausalLM.from_pretrained("lldois/v102_v74_pcgrad_normalized_task_delta", 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 lldois/v102_v74_pcgrad_normalized_task_delta with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "lldois/v102_v74_pcgrad_normalized_task_delta"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "lldois/v102_v74_pcgrad_normalized_task_delta",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/lldois/v102_v74_pcgrad_normalized_task_delta

SGLang

How to use lldois/v102_v74_pcgrad_normalized_task_delta 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 "lldois/v102_v74_pcgrad_normalized_task_delta" \
    --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": "lldois/v102_v74_pcgrad_normalized_task_delta",
		"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 "lldois/v102_v74_pcgrad_normalized_task_delta" \
        --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": "lldois/v102_v74_pcgrad_normalized_task_delta",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use lldois/v102_v74_pcgrad_normalized_task_delta with Docker Model Runner:
```
docker model run hf.co/lldois/v102_v74_pcgrad_normalized_task_delta
```

v102_v74_pcgrad_normalized_task_delta / experiment_recipe.json

lldois

Upload v102_v74_pcgrad_normalized_task_delta model without checkpoints

5b832b8 verified 10 days ago

Raw

History Blame Contribute Delete

2.05 kB

	{
	"created_at": "2026-07-27T23:14:42+08:00",
	"name": "v102_v74_pcgrad_normalized_task_delta",
	"method": "offline_parameter_delta_pcgrad",
	"method_caveat": "PCGrad is applied to specialist parameter deltas, not simultaneous minibatch gradients. This preserves the conflict-projection rule while making the experiment reproducible from existing experts.",
	"anchor": "/home/ll/llm4rec/experiments/outputs/v74_public091_rslora_loraplus_r128_a16_lr1e5_x16_ep3",
	"experts": {
	"user": "/home/ll/llm4rec/experiments/outputs/v84_public091_itemic_w2_rslora_loraplus_ep3",
	"video_ad": "/home/ll/llm4rec/experiments/outputs/v87_v74_final_itemic_focal_repair_r32_lr1e6_ep025",
	"product_live_world": "/home/ll/llm4rec/experiments/outputs/v92_public091_itemic_io_lora_plus_body_r128_ep3",
	"ad_world": "/home/ll/llm4rec/experiments/outputs/v56_public091_rslora_r128_a16_lr6e5_ep2"
	},
	"formula": "For layers 8-27 projection matrices: top-10% specialist deltas, normalize each tensor to the v87 delta norm, apply deterministic pairwise PCGrad, average at 0.25, cap at 0.30*v87 norm.",
	"changed_tensors": 140,
	"injected_global_l2": 0.005041098032500604,
	"sources_sha256": {
	"v74": "e626e97abf4a373809fa30f5909bbc44db5290cfb5d6576eb1ab37ce7165d6bc",
	"user": "98115e06e232cf1c014985b637c8411dcdcede35edb709b0d902e7ac3e769eaf",
	"video_ad": "3b310613c97d2d8af1b8f33fa7b5b8dc9b0da6b251f74355f1ac0a7fa378a5db",
	"product_live_world": "53ae3ba523780ed5f35cc72f7cc462a481480037266ea5dbe3ae6d719537f523",
	"ad_world": "ca22045c0460b7914fe1aefdda9e4f74921bd61ae5a664e1db53bd0ed3dd4bfa"
	},
	"checkpoint_policy": "Derived final model only; no intermediate checkpoints.",
	"script": "/home/ll/llm4rec/experiments/build_v102_pcgrad_task_delta.py",
	"script_sha256": "39af8c57d0a712fc0f5dad29aa130cd8985d86328e39dc425a11e151fdb6b756",
	"reproduce": "demo/LLaMA-Factory/.venv/bin/python experiments/build_v102_pcgrad_task_delta.py",
	"output_sha256": "69c70cdf66bf61cfb32890e88fe56116da952e4315c0059b019b724e61f7e7c1"
	}