Instructions to use Premchan369/Q-TensorFormer with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Premchan369/Q-TensorFormer with Transformers:

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

pipe = pipeline("text-generation", model="Premchan369/Q-TensorFormer", trust_remote_code=True)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Premchan369/Q-TensorFormer", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Premchan369/Q-TensorFormer with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Premchan369/Q-TensorFormer"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Premchan369/Q-TensorFormer",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Premchan369/Q-TensorFormer

SGLang

How to use Premchan369/Q-TensorFormer 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 "Premchan369/Q-TensorFormer" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Premchan369/Q-TensorFormer",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "Premchan369/Q-TensorFormer" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Premchan369/Q-TensorFormer",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Premchan369/Q-TensorFormer with Docker Model Runner:
```
docker model run hf.co/Premchan369/Q-TensorFormer
```

Q-TensorFormer / results /CHECKPOINT_MODEL_MATCH.md

Premchandyadav369

feat(hf): turnkey custom model with TensorTrainLinear, AdaptiveKVCache, PID controller, NVML profiler, OpenAI server, and drop-in compressor

15a4381 25 days ago

preview code

Raw History Blame Contribute Delete

5.77 kB

Checkpoint-to-Model Diagnostic Matching Matrix

1. Executive Summary

This report delivers a forensic audit and diagnostic loading matrix of the checkpoint file qtensorformer_small.pt against all candidate model architectures in the repository. Under the scientific mandate strict=True, zero candidates achieve checkpoint compatibility.

2. Status Distinctions

PROVEN FACT

qtensorformer_small.pt is a serialized PyTorch checkpoint dictionary containing 6 top-level keys:
- model_state_dict (113 tensor weights, total 1,347,747 parameters)
- config ({'vocab_size': 10000, 'hidden_dim': 128, 'intermediate_size': 256, 'n_heads': 4, 'n_layers': 3, 'seq_len': 64, 'tt_rank': 4, 'n_qubits': 4, 'quantum_layers': 2, 'quantum_ratio': 0.3, 'use_quantum_attention': True, 'use_adaptive_rank': True, 'use_selective_routing': True, 'r_min': 2, 'r_max': 12, 'alpha': 1.0})
- params: 1347747
- baseline_params: 16146432
- compression_ratio: 11.980313812607262
- version: "1.0.0"
The state dictionary relies on TensorLy tensor decomposition layers:
- blocks.i.ffn.layer.weight.factors.factor_0..5
- blocks.i.ffn2.layer.weight.factors.factor_0..5
- lm_head.layer.weight.factors.factor_0..5
- embeddings.quantum_enhancer.project_up.weight
- blocks.i.attention.quantum_encoder.project_up.weight
- blocks.i.attention.alpha_gate.0.weight
Across all candidate model implementations in the repository:
- Common keys with qtensorformer_small.pt: 0
- Unexpected keys from qtensorformer_small.pt: 113 (100% of checkpoint keys)
- Strict load result: FAIL on 100% of candidate models.
Git commit history proves qtensorformer_small.pt was committed in commit 82fd973cb44f62a21ac0fc0fdfa1b30a8c598754 on May 6, 2026 by ml-intern, originating from an external /app/qtensorformer container package.

INFERENCE

The model architecture that produced qtensorformer_small.pt was developed in an ephemeral Docker environment and was never added to the git repository source tree.
Any attempt to load qtensorformer_small.pt into any repository class would require strict=False, which would silently discard all 113 trained weight tensors and run random uninitialized weights, violating scientific integrity.

UNRESOLVED

The original source repository of the /app/qtensorformer package that created qtensorformer_small.pt cannot be retrieved from git history.

3. Checkpoint Forensic Specification

Attribute	Checkpoint Value
File Name	`qtensorformer_small.pt`
File Size	5,432,305 bytes
Object Type	`dict`
Top-Level Keys	`['model_state_dict', 'config', 'params', 'baseline_params', 'compression_ratio', 'version']`
Parameter Count	1,347,747
Tensor Count	113
Vocabulary Dimension	10,000 (`embeddings.token_embedding.weight`: `[10000, 128]`)
Hidden Dimension ($d_{\text{model}}$)	128
Number of Layers	3 (`blocks.0`, `blocks.1`, `blocks.2`)
Attention Projection	`q_proj`: `[4, 32]`, `classical_proj`: `[128, 128]`
FFN Factorization	TensorLy TT Factorized (`factor_0..5` for `ffn` and `ffn2`)
Language Model Head	TensorLy TT Factorized (`lm_head.layer.weight.factors.factor_0..5`)
Positional Encoding Length	64 (`embeddings.position_embedding.weight`: `[64, 128]`)

4. Candidate Model Load Matrix

Candidate Model	Config Class Used	Config Construction	Model Construction	`strict=True` Load	Missing Keys	Unexpected Keys	Shape Mismatches	Param Count Match
A1: `q_tensor_former.QTensorFormer` (Default)	`q_tensor_former.CFG()`	PASS	PASS	FAIL	69	113	N/A (0 key overlap)	115,428 vs 1,347,747 (No)
A2: `q_tensor_former.QTensorFormer` (Matched Dims)	`q_tensor_former.CFG(vocab_size=10000, d_model=128, n_layers=3, max_seq_len=64)`	PASS	PASS	FAIL	101	113	N/A (0 key overlap)	1,347,747 vs 1,347,747 (No)
B1: `src.models.QTensorFormer` (Default)	`src.config.ModelConfig()`	PASS	PASS	FAIL	125	113	N/A (0 key overlap)	1,460,860 vs 1,347,747 (No)
B2: `src.models.QTensorFormer` (Matched Dims)	`src.config.ModelConfig(vocab_size=10000, d_model=128, n_layers=3, max_seq_len=64)`	PASS	PASS	FAIL	185	113	N/A (0 key overlap)	1,526,428 vs 1,347,747 (No)
C1: `q_tensor_former_v2.QTensorFormer` (Default)	`q_tensor_former_v2.Config()`	PASS	PASS	FAIL	73	113	N/A (0 key overlap)	1,476,004 vs 1,347,747 (No)
C2: `q_tensor_former_v2.QTensorFormer` (Matched Dims)	`q_tensor_former_v2.Config(vocab_size=10000, d_model=128, n_layers=3, max_seq_len=64)`	PASS	PASS	FAIL	107	113	N/A (0 key overlap)	1,542,044 vs 1,347,747 (No)
D1: `src.hf_model.QTensorFormerForCausalLM`	`src.hf_model.QTensorFormerConfig()`	PASS	PASS	FAIL	125	113	N/A (0 key overlap)	1,460,860 vs 1,347,747 (No)

5. Failure Diagnosis & Scientific Conclusion

Exact Error Raised:

RuntimeError: Error(s) in loading state_dict for QTensorFormer:
    Missing key(s) in state_dict: ...
    Unexpected key(s) in state_dict: "embeddings.token_embedding.weight", ...

Scientific Invariant: Under the rule NO RESULT > INVALID RESULT, loading weights with strict=False or synthesizing dummy modules is strictly forbidden.
Verdict: CHECKPOINT_STRICT_LOAD: FAIL for all candidate models in the repository.