Promote refract to top-level package (refract/, not turboquant/refract/)

REFRACT is meant to be a general KV-quant quality oracle, applicable
across model families (gemma, qwen, mistral, phi, llama, etc.) and not
TurboQuant-specific. Top-level placement reflects that scope and gives
a cleaner CLI: `python3 -m refract.cli` instead of nested under turboquant.

Updates:
- git mv turboquant/refract/ -> refract/
- All `from turboquant.refract` -> `from refract` imports rewritten
- pyproject.toml: include both turboquant* and refract* in setuptools
- All "turboquant/refract/" path references in README/tests updated

Verified: 14/14 unit tests pass, CLI module path works.

pyproject.toml

@@ -18,7 +18,7 @@ bench = [
 ]
 
 [tool.setuptools.packages.find]
-include = ["turboquant*"]
+include = ["turboquant*", "refract*"]
 
 [build-system]
 requires = ["setuptools>=68"]

refract/README.md

@@ -55,11 +55,11 @@ export LLAMA_CPP_BIN_DIR=/path/to/llama.cpp/build/bin
 ## Quickstart
 
 ```bash
-python3 -m turboquant.refract.cli score \
+python3 -m refract.cli score \
     --model ~/local_llms/models/Qwen2.5-7B-Instruct-Q8_0.gguf \
     --reference 'ctk=f16,ctv=f16' \
     --candidate 'ctk=q8_0,ctv=turbo4,attn_rot_v=0' \
-    --prompts turboquant/refract/prompts/v0.1.jsonl \
+    --prompts refract/prompts/v0.1.jsonl \
     --corpus ~/local_llms/llama.cpp/wikitext-2-raw/wiki.test.raw \
     --chunks 32 -c 512 -ngl 99 \
     --measure-floor \
@@ -143,17 +143,17 @@ It is marked `integration` because it spawns llama.cpp many times against a
 ```bash
 # Mark it in pytest.ini if not already:
 #   markers = integration: marks tests requiring llama.cpp + a real GGUF
-pytest -m integration turboquant/refract/tests/test_validation.py -s
+pytest -m integration refract/tests/test_validation.py -s
 ```
 
 Or directly via CLI:
 
 ```bash
-python3 -m turboquant.refract.cli score \
+python3 -m refract.cli score \
     --model ~/local_llms/models/gemma-4-26B-A4B-Q8_0.gguf \
     --reference 'ctk=f16,ctv=f16' \
     --candidate 'ctk=q8_0,ctv=turbo4,attn_rot_v=0,attn_rot_k=0' \
-    --prompts turboquant/refract/prompts/v0.1.jsonl \
+    --prompts refract/prompts/v0.1.jsonl \
     --corpus ~/local_llms/llama.cpp/wikitext-2-raw/wiki.test.raw \
     --chunks 32 --measure-floor
 ```
@@ -163,7 +163,7 @@ Fast unit tests (no subprocess) live in `tests/test_unit.py`.
 ## File layout
 
 ```
-turboquant/refract/
+refract/
   __init__.py
   cli.py                    # CLI entry point
   score.py                  # composite + bands + floor logic

refract/__init__.py

@@ -9,7 +9,7 @@ quantization quality. Replaces "lower PPL = better" with a composite of:
     (v0.1 uses corpus KLD via llama-perplexity as a proxy for trajectory KLD).
 
 See:
-  - turboquant/refract/README.md for usage.
+  - refract/README.md for usage.
   - docs/papers/attn-rotation-and-ppl-artifact.md for the motivating paper.
 """
 

refract/cli.py

@@ -1,11 +1,11 @@
 """REFRACT v0.1 CLI.
 
 Usage:
-    python3 -m turboquant.refract.cli score \\
+    python3 -m refract.cli score \\
         --model MODEL.gguf \\
         --reference "ctk=f16,ctv=f16" \\
         --candidate "ctk=q8_0,ctv=turbo4,attn_rot_v=0" \\
-        --prompts turboquant/refract/prompts/v0.1.jsonl \\
+        --prompts refract/prompts/v0.1.jsonl \\
         --corpus ~/local_llms/llama.cpp/wikitext-2-raw/wiki.test.raw \\
         --chunks 32 -c 512 -ngl 99
 """

refract/tests/test_unit.py

@@ -1,7 +1,7 @@
 """Fast unit tests for REFRACT scoring math + parsers. No subprocess.
 
 Run with:
-    pytest turboquant/refract/tests/test_unit.py
+    pytest refract/tests/test_unit.py
 """
 
 from __future__ import annotations
@@ -10,10 +10,10 @@ import math
 
 import pytest
 
-from turboquant.refract.axes.gtm import _diff, _tokenize_words
-from turboquant.refract.axes.kld import _kld_to_score
-from turboquant.refract.runner import KVConfig
-from turboquant.refract.score import (
+from refract.axes.gtm import _diff, _tokenize_words
+from refract.axes.kld import _kld_to_score
+from refract.runner import KVConfig
+from refract.score import (
     MIN_FLOOR,
     band,
     composite_score,

refract/tests/test_validation.py

@@ -20,15 +20,15 @@ must NOT mistake that for a pass.
 This test is integration-only: it spawns llama-cli + llama-perplexity many
 times against a 26B-A4B model and takes 30+ minutes. Run with:
 
-    pytest -m integration turboquant/refract/tests/test_validation.py
+    pytest -m integration refract/tests/test_validation.py
 
 or directly:
 
-    python3 -m turboquant.refract.cli score \\
+    python3 -m refract.cli score \\
         --model ~/local_llms/models/gemma-4-26B-A4B-Q8_0.gguf \\
         --reference 'ctk=f16,ctv=f16' \\
         --candidate 'ctk=q8_0,ctv=turbo4,attn_rot_v=0,attn_rot_k=0' \\
-        --prompts turboquant/refract/prompts/v0.1.jsonl \\
+        --prompts refract/prompts/v0.1.jsonl \\
         --corpus ~/local_llms/llama.cpp/wikitext-2-raw/wiki.test.raw \\
         --chunks 32 --measure-floor
 """
@@ -40,10 +40,10 @@ from pathlib import Path
 
 import pytest
 
-from turboquant.refract.axes.gtm import run_gtm
-from turboquant.refract.axes.kld import run_kld
-from turboquant.refract.runner import KVConfig
-from turboquant.refract.score import composite_score
+from refract.axes.gtm import run_gtm
+from refract.axes.kld import run_kld
+from refract.runner import KVConfig
+from refract.score import composite_score
 
 
 # Default paths; override via env if your model lives elsewhere.