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 / GENERATION_VALIDATION_REPORT.md

Premchandyadav369

fix(config): Align configuration API with canonical ModelConfig fields and document checkpoint provenance

0b7ee79 about 2 months ago

preview code

Raw History Blame Contribute Delete

2.39 kB

# GENERATION VALIDATION REPORT: Autoregressive Decoding Audit

**Date**: September 17, 2026  
**Auditor**: Senior ML Systems Engineer & Research Software Auditor  
**Final Status**: **`GENERATION_PASS`**
---
## 1. Generation Interface Verification
The repository provides generation interfaces at two levels:
1. **Model-Level Method**: `QTensorFormer.generate()` in `src/models.py` (with `AdaptiveKVCache`).
2. **Hugging Face Integration**: `QTensorFormerForCausalLM` in `src/hf_model.py` (inherits `transformers.GenerationMixin`).
3. **Pure PyTorch Greedy Loop**: Evaluated via standard causal next-token prediction `argmax(logits[:, -1, :])` directly on the model's `forward()` output contract.
---
## 2. Deterministic Reproducibility Audit
Deterministic greedy decoding was executed twice from identical initial input prompt `[10, 25, 30]` with `seed=42`:
### Test Run Comparison
- **`src.models.QTensorFormer`**:
  - Run 1 Tokens: `[10, 25, 30, 5278, 5278, 5278, 5278, 5278]`
  - Run 2 Tokens: `[10, 25, 30, 5278, 5278, 5278, 5278, 5278]`
  - Match: **100% Identical (Zero divergence)**

- **`q_tensor_former.QTensorFormer`**:
  - Run 1 Tokens: `[10, 25, 30, 2856, 1538, 2498, 2498, 2498]`
  - Run 2 Tokens: `[10, 25, 30, 2856, 1538, 2498, 2498, 2498]`
  - Match: **100% Identical (Zero divergence)**

---

## 3. Per-Token Sequence Growth & Dimension Trace

| Step | Current Sequence Length ($T$) | Logits Shape | Selected Next Token ID | Next Sequence Length ($T+1$) | Invariant Check |
| :---: | :---: | :---: | :---: | :---: | :---: |
| Prompt | 3 | N/A | N/A | 3 | Initial prompt |
| 1 | 3 | `[1, 3, 10000]` | $5278$ | 4 | **VALID** |
| 2 | 4 | `[1, 4, 10000]` | $5278$ | 5 | **VALID** |
| 3 | 5 | `[1, 5, 10000]` | $5278$ | 6 | **VALID** |
| 4 | 6 | `[1, 6, 10000]` | $5278$ | 7 | **VALID** |
| 5 | 7 | `[1, 7, 10000]` | $5278$ | 8 | **VALID** |

---

## 4. Verification Checklist
- [x] Input IDs semantics: Int64 tensor indexing within $[0, V-1]$.
- [x] Logits semantics: Real-valued unnormalized log-probabilities over vocabulary.
- [x] Next-token selection: Greedy argmax along vocabulary dimension.
- [x] Sequence growth: Monotonic single-token concatenation.
- [x] Determinism: Repeated execution with identical prompt and seed yields identical sequences.
- [x] Numerical stability: No NaN or Inf encountered during unrolling.

**FINAL STATUS**: **`GENERATION_PASS`**

	# GENERATION VALIDATION REPORT: Autoregressive Decoding Audit

	Date: September 17, 2026
	Auditor: Senior ML Systems Engineer & Research Software Auditor
	Final Status: `GENERATION_PASS`

	---

	## 1. Generation Interface Verification

	The repository provides generation interfaces at two levels:
	1. Model-Level Method: `QTensorFormer.generate()` in `src/models.py` (with `AdaptiveKVCache`).
	2. Hugging Face Integration: `QTensorFormerForCausalLM` in `src/hf_model.py` (inherits `transformers.GenerationMixin`).
	3. Pure PyTorch Greedy Loop: Evaluated via standard causal next-token prediction `argmax(logits[:, -1, :])` directly on the model's `forward()` output contract.

	---

	## 2. Deterministic Reproducibility Audit

	Deterministic greedy decoding was executed twice from identical initial input prompt `[10, 25, 30]` with `seed=42`:

	### Test Run Comparison
	- `src.models.QTensorFormer`:
	- Run 1 Tokens: `[10, 25, 30, 5278, 5278, 5278, 5278, 5278]`
	- Run 2 Tokens: `[10, 25, 30, 5278, 5278, 5278, 5278, 5278]`
	- Match: 100% Identical (Zero divergence)

	- `q_tensor_former.QTensorFormer`:
	- Run 1 Tokens: `[10, 25, 30, 2856, 1538, 2498, 2498, 2498]`
	- Run 2 Tokens: `[10, 25, 30, 2856, 1538, 2498, 2498, 2498]`
	- Match: 100% Identical (Zero divergence)

	---

	## 3. Per-Token Sequence Growth & Dimension Trace

	\| Step \| Current Sequence Length ($T$) \| Logits Shape \| Selected Next Token ID \| Next Sequence Length ($T+1$) \| Invariant Check \|
	\| :---: \| :---: \| :---: \| :---: \| :---: \| :---: \|
	\| Prompt \| 3 \| N/A \| N/A \| 3 \| Initial prompt \|
	\| 1 \| 3 \| `[1, 3, 10000]` \| $5278$ \| 4 \| VALID \|
	\| 2 \| 4 \| `[1, 4, 10000]` \| $5278$ \| 5 \| VALID \|
	\| 3 \| 5 \| `[1, 5, 10000]` \| $5278$ \| 6 \| VALID \|
	\| 4 \| 6 \| `[1, 6, 10000]` \| $5278$ \| 7 \| VALID \|
	\| 5 \| 7 \| `[1, 7, 10000]` \| $5278$ \| 8 \| VALID \|

	---

	## 4. Verification Checklist
	- [x] Input IDs semantics: Int64 tensor indexing within $[0, V-1]$.
	- [x] Logits semantics: Real-valued unnormalized log-probabilities over vocabulary.
	- [x] Next-token selection: Greedy argmax along vocabulary dimension.
	- [x] Sequence growth: Monotonic single-token concatenation.
	- [x] Determinism: Repeated execution with identical prompt and seed yields identical sequences.
	- [x] Numerical stability: No NaN or Inf encountered during unrolling.

	FINAL STATUS: `GENERATION_PASS`