Update links after username change
Browse files
README.md
CHANGED
|
@@ -12,7 +12,7 @@ I build practical, inspectable tools around open models: adapters, quantization
|
|
| 12 |
|
| 13 |
## Featured work
|
| 14 |
|
| 15 |
-
### [PathPack-Q · LFM2.5-2.6B training-free quantization](https://huggingface.co/
|
| 16 |
|
| 17 |
An architecture-specific post-training quantization experiment for LiquidAI's
|
| 18 |
hybrid convolution/attention LLM. PathPack-Q uses exact gated-path channel
|
|
@@ -24,7 +24,7 @@ without training, text calibration data, extra parameters, or runtime operators.
|
|
| 24 |
- **Identical 1,517,616,892-byte checkpoint** and 4.501 effective bits/weight
|
| 25 |
- Includes the search algorithm, complete-path acceptance gate, rejected-layer evidence, checkpoint builder, and machine-readable evaluations
|
| 26 |
|
| 27 |
-
### [ScopeGuard · Qwen2.5-1.5B LoRA](https://huggingface.co/
|
| 28 |
|
| 29 |
A locally trained agent decision-layer adapter that turns natural-language requests into strict JSON risk and confirmation decisions before tools execute.
|
| 30 |
|
|
@@ -33,15 +33,15 @@ A locally trained agent decision-layer adapter that turns natural-language reque
|
|
| 33 |
- 3.957M trainable parameters — only 0.256% of the 1.5B base model
|
| 34 |
- Includes adapter weights, original dataset, deterministic generator, training config, baseline outputs, and per-example evaluation
|
| 35 |
|
| 36 |
-
→ [Explore the ScopeGuard dataset](https://huggingface.co/datasets/
|
| 37 |
|
| 38 |
-
→ [Open the complete benchmark explorer](https://huggingface.co/spaces/
|
| 39 |
|
| 40 |
-
### [LoRA Lens](https://huggingface.co/spaces/
|
| 41 |
|
| 42 |
An in-browser audit tool for `adapter_config.json` files. It surfaces rank, alpha, scaling, target modules, reproducibility gaps, and conservative parameter-efficiency estimates without uploading weights or requiring an API key.
|
| 43 |
|
| 44 |
-
→ [Inspect the implementation](https://huggingface.co/
|
| 45 |
|
| 46 |
## Current lab
|
| 47 |
|
|
|
|
| 12 |
|
| 13 |
## Featured work
|
| 14 |
|
| 15 |
+
### [PathPack-Q · LFM2.5-2.6B training-free quantization](https://huggingface.co/praveenkumarpranjal/LFM2.5-2.6B-4bit-PathPack-Q)
|
| 16 |
|
| 17 |
An architecture-specific post-training quantization experiment for LiquidAI's
|
| 18 |
hybrid convolution/attention LLM. PathPack-Q uses exact gated-path channel
|
|
|
|
| 24 |
- **Identical 1,517,616,892-byte checkpoint** and 4.501 effective bits/weight
|
| 25 |
- Includes the search algorithm, complete-path acceptance gate, rejected-layer evidence, checkpoint builder, and machine-readable evaluations
|
| 26 |
|
| 27 |
+
### [ScopeGuard · Qwen2.5-1.5B LoRA](https://huggingface.co/praveenkumarpranjal/scopeguard-qwen2.5-1.5b-lora)
|
| 28 |
|
| 29 |
A locally trained agent decision-layer adapter that turns natural-language requests into strict JSON risk and confirmation decisions before tools execute.
|
| 30 |
|
|
|
|
| 33 |
- 3.957M trainable parameters — only 0.256% of the 1.5B base model
|
| 34 |
- Includes adapter weights, original dataset, deterministic generator, training config, baseline outputs, and per-example evaluation
|
| 35 |
|
| 36 |
+
→ [Explore the ScopeGuard dataset](https://huggingface.co/datasets/praveenkumarpranjal/scopeguard-decisions)
|
| 37 |
|
| 38 |
+
→ [Open the complete benchmark explorer](https://huggingface.co/spaces/praveenkumarpranjal/scopeguard-benchmark)
|
| 39 |
|
| 40 |
+
### [LoRA Lens](https://huggingface.co/spaces/praveenkumarpranjal/lora-lens-demo)
|
| 41 |
|
| 42 |
An in-browser audit tool for `adapter_config.json` files. It surfaces rank, alpha, scaling, target modules, reproducibility gaps, and conservative parameter-efficiency estimates without uploading weights or requiring an API key.
|
| 43 |
|
| 44 |
+
→ [Inspect the implementation](https://huggingface.co/praveenkumarpranjal/lora-lens)
|
| 45 |
|
| 46 |
## Current lab
|
| 47 |
|