Post-Training Quantization

This directory holds the post-training quantization passes used to shrink the fine-tuned Qwen Eliza checkpoints before they leave the training rig. Each method is a self-contained CLI; they are independent and can be combined or compared on the same fine-tuned checkpoint.

PolarQuant

Reference: Caio Vicentino, PolarQuant: Optimal Gaussian Weight Quantization via Hadamard Rotation for LLM Compression, arXiv:2603.29078 (March 2026). The arXiv PDF was withdrawn for errata fixes; the implementation we run against lives in caiovicentino/eoq-quantization @ commit 15a12160245d7d3015290c6c5b6dbb7f22094d5e. The two source files we depend on are vendored under polarquant/ (see polarquant/LICENSE.md for the upstream license situation — currently no LICENSE file in the upstream repo; vendored under a documented research-citation arrangement that must be revisited if the upstream project publishes a restrictive license).

Algorithm in one paragraph

For each linear's weight tensor, group the values into power-of-two blocks (default 128). Per block: extract the L2 norm and store it as fp16; normalize the block to the unit hypersphere; multiply by a Walsh–Hadamard matrix so that each coordinate is an i.i.d. draw from roughly N(0, 1/√d); quantize each coordinate against the Lloyd–Max optimal centroids for N(0, 1) (precomputed via the iterative conditional-expectation algorithm in polarquant/polar_quant.py), storing the centroid index as int8; optionally append a 1-bit QJL residual sign per block as a cheap correction. At inference, undo each step: lookup centroid → divide by √d → inverse Hadamard → rescale by per-block norm. The Hadamard rotation alone accounts for ~98% of the perplexity recovery vs absmax (paper, Ablation §4.2). PolarQuant is data-free — no calibration set is required — because the post-rotation distribution is analytically Gaussian.

Tradeoffs

• Pros. Data-free; near-lossless at Q5 (paper claims very small PPL deltas on Qwen-class checkpoints vs FP16). int8 codes + fp16 per-block norms gives the storage payload that downstream INT4 inference kernels (torchao, llama.cpp, MLX) consume directly. Architecture-agnostic at the nn.Linear level.
• Cons. Reconstruction-back-into-fp16 (the path we use today) does not save inference VRAM by itself — the model still loads as a fp16 nn.Linear matrix; the win shows up when the codes-only payload is loaded by an INT4 inference kernel. The sidecar polarquant_artifacts.safetensors we emit is the artifact for that downstream path. The vendored compute kernel is a pure-Python PyTorch loop over linears (no Triton, no fused kernel), so the quantization step itself is slow on big models — it costs O(N) Hadamard products at quantization time, then is free at inference.

Supported architectures

The vendored kernel runs on any model that exposes its weights as nn.Linear modules. We have explicitly verified it on:

• Qwen2 / Qwen2.5 (Qwen2ForCausalLM)
• Qwen3 (Qwen3ForCausalLM)
• Llama, Mistral, Phi-3 (the linear layout matches Qwen2; should work)

Qwen3.5 caveat (read this)

The active targets Qwen3.5-{0.8B, 2B, 4B} ship as multimodal hybrid-attention checkpoints (Qwen3_5ForConditionalGeneration). The text decoder mixes linear_attention and full_attention layers and adds Mamba-style SSM state, mrope vision embeddings, and a separate vision encoder. PolarQuant operates on nn.Linear weights, so it will still quantize the Q/K/V/O and MLP projections inside both layer types — but:

1. The Mamba-style SSM in linear_attention layers carries a few non-linear state buffers (A_log, D, dt_bias) that are not nn.Linear. We deliberately skip those (they fall under the min_numel cutoff or aren't nn.Linear instances) — quantizing them through PolarQuant's Gaussian assumption would produce garbage.
2. The default AutoModelForCausalLM loader will refuse Qwen3.5 since the architecture is ForConditionalGeneration. The orchestrator needs to load it via AutoModelForVision2Seq (or the Qwen3_5ForConditionalGeneration class directly) and walk the text decoder. We have not done that integration yet — for the 0.8B validation we fall back to Qwen/Qwen3.5-0.8B.
3. Future MoE variants have router weights — those are tiny, fall under the --min-numel cutoff, and must be deliberately skipped when that line is reintroduced.

If/when we need to ship PolarQuant on a Qwen3.5 checkpoint, the fix is: load with the right AutoModel* class, expose the text decoder via model.language_model (or whichever attribute carries the text tower), then call quantize_checkpoint against that submodule. The rest of the kernel does not need to change.

CLI

Quantize a fine-tuned 2B checkpoint:

uv run python scripts/quantization/polarquant_apply.py \
    --model checkpoints/qwen35-2b-eliza/final \
    --calibration data/final/val.jsonl \
    --calibration-samples 128 \
    --output checkpoints/qwen35-2b-eliza/final-polarquant

The --calibration* flags are accepted for parity with the rest of the quantization pipeline but are unused — PolarQuant is data-free per the paper. Passing a non-existent calibration path still errors so that misconfigured callers fail loudly.

Useful knobs:

• --bits {2,3,4,5,6} (default 4). Q5 is the paper's near-lossless point; Q4 is the practical default for INT4 inference kernels.
• --block-size N (default 128, must be a power of 2).
• --no-qjl to drop the 1-bit residual correction (slightly worse PPL, slightly smaller sidecar).
• --no-artifacts to skip writing the sidecar codes; useful when you only want the reconstructed fp16 model.
• --include-lm-head / --include-embedding to override the default skip behavior. Quantizing the LM head on weight-tied models is redundant; quantizing embedding tables hurts logits because they're looked up rather than multiplied.

Validation

scripts/quantization/test_polarquant.py runs the round-trip on Qwen/Qwen3.5-0.8B (the closest text-only causal-LM stand-in for Qwen/Qwen3.5-0.8B — see caveat above), using 5 native JSON-shaped samples from data/final/val.jsonl. It asserts (a) the codes-only payload is at least 30% smaller than the fp16 baseline checkpoint and (b) the quantized model produces non-degenerate text on every sample.

uv run python scripts/quantization/test_polarquant.py

TurboQuant

Reference: Amir Zandieh, Majid Daliri, Majid Hadian, Vahab Mirrokni (Google Research / Google DeepMind / NYU), TurboQuant: Online Vector Quantization with Near-Optimal Distortion Rate, arXiv:2504.19874, ICLR 2026. Implementation: vivekvar-dl/turboquant shipped to PyPI as turbokv v0.1.0 (import name: turboquant). Pinned in pyproject.toml under [project.optional-dependencies] train.

What this is — and what it is NOT

TurboQuant is an online, data-oblivious KV-cache quantizer. It runs at inference time and replaces the standard fp16 DynamicCache with a TurboQuantCache that stores keys and values in 2- or 4-bit packed form per coordinate.

It is not a weight quantizer. The model.safetensors file does not change. turboquant_apply.py therefore writes the merged base model unchanged and adds a turboquant.json sidecar that records the quantizer config (nbits, base_seed, calibrated skip_layers, residual_length) so a downstream loader can reconstruct the cache deterministically.

Algorithm in one paragraph

For each cached vector x of length head_dim (per attention head, per token): (1) extract and store ||x|| in bf16; (2) rotate the unit vector by a fixed-per-layer random orthogonal matrix Π drawn from SO(d) via QR-decomposition with sign-fix — after rotation each coordinate is approximately Beta-distributed (≈ N(0, 1/d) for large d); (3) bucketize each coordinate against the precomputed Lloyd-Max optimal centroids for the Beta distribution at b bits; (4) bit-pack the indices into uint8 (2 indices/byte at 4-bit, 4 indices/byte at 2-bit). At attention time, unpack → centroid lookup → inverse rotation x ≈ ||x|| · ΠᵀC[i]. The recent residual_length tokens stay in fp16 to keep the freshly-generated context lossless.

Tradeoffs

• Pros. Data-free / online — calibration is a single forward pass used only to detect outlier-norm layers (typically only layer 0) that should stay fp16. Drops naturally into model.generate via past_key_values=cache. Works across architectures (Qwen, Llama, Gemma, Phi) without per-model code paths. Information-theoretic near-optimal: paper proves the rate is within ~2.7× the per-channel Shannon-Bennett lower bound.
• Cons. The reference implementation is pure PyTorch — the per-step quantize/dequantize is a Python-level operation per layer per step, which costs throughput. On a 0.8B model on a 5080 we observed ~5× slowdown vs the bf16 DynamicCache (66.8 → 12.2 tok/s). The TurboQuant paper claims faster runtime than the bf16 baseline because it ships Triton kernels; those are not in the turbokv 0.1.0 PyPI release we depend on. Until upstream ships Triton, this method is a memory win, not a speed win.
• The savings are concentrated in the long-context regime. At 4096- token prefill on Qwen3.5-0.8B we measured 3.52× per-token KV reduction (114,688 → 32,608 bytes/token) which produced a real 274 MB peak-VRAM drop on a tiny model — the absolute savings scale with num_hidden_layers × num_kv_heads × head_dim × context_length.

Supported architectures

TurboQuantCache materializes a TurboQuantLayer per full-attention layer reported by the model config. Verified locally against:

• Qwen3 (Qwen3ForCausalLM) — single-mode full attention. Works.

Should work, by structural inspection, on:

• Qwen2 / Qwen2.5, Llama, Gemma, Phi — all uniform full-attention with GQA, the same shape TurboQuantLayer already handles.

Qwen3.5 caveat (read this)

The active targets Qwen3.5-{0.8B, 2B, 4B} are **hybrid linear-attention

• Gated Attention** models (Qwen3_5ForConditionalGeneration). The text decoder declares per-layer layer_types: most layers are linear_attention (Gated DeltaNet — recurrent state, no (B, H, T, D) KV cache) and only a minority (typically every 4th layer) are full_attention. TurboQuant is only meaningful for the full-attention layers — there is nothing to quantize in a recurrent state. Concretely on Qwen3.5-2B, 6 of 24 layers are full attention; the analytic ceiling on KV reduction is therefore at most ~25% of the standard ratio applied to those 6 layers. The kv_bytes_per_token_analytic helper in test_turboquant.py honors layer_types so the reported reduction factor is correct for hybrid models.

The 0.8B target is also a vision-language checkpoint (Qwen3_5ForConditionalGeneration with vision encoder), so loading it needs AutoModelForVision2Seq (or the Qwen3_5ForConditionalGeneration class directly), and TurboQuantCache(model.config, ...) must receive the text decoder config — model.config.get_text_config(decoder=True). The cache.py in turbokv 0.1.0 already calls get_text_config when available, so this should work in principle, but we have not validated it end-to-end on a Qwen3.5 checkpoint yet — for the 0.8B-class validation we fall back to Qwen/Qwen3.5-0.8B.

For future dense/MoE variants, TurboQuant is orthogonal to expert routing when the KV cache shape is unchanged. Revalidate that separately before adding those tiers back to the active release line.

CLI

Apply to a fine-tuned 2B checkpoint (auto-merges if --model points to a LoRA adapter):

uv run python scripts/quantization/turboquant_apply.py \
    --model checkpoints/qwen35-2b-eliza/final \
    --calibration data/final/val.jsonl \
    --calibration-samples 128 \
    --output checkpoints/qwen35-2b-eliza/final-turboquant

Useful knobs:

• --nbits {2,4} (default 4). 4-bit hits ~3.5× KV reduction with ~zero quality loss; 2-bit hits ~6.4× with measurable but small degradation per the paper.
• --residual-length N (default 128). Most-recent N tokens stay in fp16 to keep freshly-generated context lossless.
• --base-seed N (default 42). Layer i uses seed = base_seed + i. Pin this to the value recorded in turboquant.json at inference time; otherwise the rotation matrices will not match and dequant gives garbage.
• --norm-threshold f (default 5.0). Calibration skips layers whose per-token key norms exceed f × median.

Inference-time use

import json
from turboquant import TurboQuantCache
from transformers import AutoModelForCausalLM, AutoTokenizer

side = json.load(open("checkpoints/qwen35-2b-eliza/final-turboquant/turboquant.json"))
model = AutoModelForCausalLM.from_pretrained(
    "checkpoints/qwen35-2b-eliza/final-turboquant",
    torch_dtype="bfloat16", device_map="cuda",
)
tok = AutoTokenizer.from_pretrained("checkpoints/qwen35-2b-eliza/final-turboquant")

cache = TurboQuantCache(
    model.config,
    nbits=side["nbits"],
    residual_length=side["residual_length"],
    base_seed=side["base_seed"],
    skip_layers=set(side["skip_layers"]),
)
out = model.generate(**tok("...", return_tensors="pt").to("cuda"),
                     past_key_values=cache, max_new_tokens=256)

Validation

scripts/quantization/test_turboquant.py runs the round-trip on Qwen/Qwen3.5-0.8B (closest text-only stand-in for Qwen/Qwen3.5-0.8B — see caveat above), with 5 native JSON-shaped prompts from data/final/val.jsonl and a 4096-token long-context probe. It asserts (a) the per-token KV-cache size shrinks by at least 30% and (b) every quantized output is non-empty and not degenerate.

uv run python scripts/quantization/test_turboquant.py

The full numeric report is written to scripts/quantization/turboquant_report.json. Last measured run on Qwen3.5-0.8B / 5080 (4-bit, skip={0}, 4096-token long context):

Fused TurboQuant (Triton)

Reference: same paper as TurboQuant above (arXiv:2504.19874). Implementation: fused-turboquant v0.1.0 — a re-implementation of the TurboQuant scheme in Triton kernels (encode, decode, fused Q@K^T scoring directly from packed indices). The math is identical to turbokv 0.1.0 above; the win is throughput.

Pinned in pyproject.toml under [project.optional-dependencies] train alongside turbokv. The two are kept side-by-side because (a) turbokv is the architecture-portable fallback when patch_model rejects a model (non-power-of-2 head_dim, fused QKV, sliding window), and (b) fused-turboquant needs a working Triton + system-dev-headers stack that turbokv does not.

CLI

Apply to a fine-tuned 27B checkpoint (auto-merges if --model points to a LoRA adapter):

uv run python scripts/quantization/fused_turboquant_apply.py \
    --model checkpoints/qwen35-4b-eliza/final \
    --output checkpoints/qwen35-4b-eliza/final-fused-turboquant \
    --bits 4

The script runs check_model_compatibility(model) first and refuses to proceed if the architecture isn't supported (logged with a structured report). After patching it discards the cache, unpatches the model, and saves the unmodified base weights — fused-TurboQuant is a runtime concern, the safetensors files are byte-identical to the input.

--calibration / --calibration-samples are accepted for parity with the other quantizer CLIs but are unused: the Lloyd-Max codebooks and RHT seeds are data-oblivious.

Inference-time use

from quantization.fused_turboquant_vendored.hf import patch_model
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "checkpoints/qwen35-4b-eliza/final-fused-turboquant",
    torch_dtype="bfloat16", device_map="cuda",
)
tok = AutoTokenizer.from_pretrained("checkpoints/qwen35-4b-eliza/final-fused-turboquant")

cache = patch_model(model, bits=4, compress_v=True)  # patches model.forward in-place
out = model.generate(**tok("...", return_tensors="pt").to("cuda"),
                     past_key_values=cache, use_cache=True, max_new_tokens=256)

Compatibility constraints (read this before deploying)

The Triton-kernel path is more constrained than the pure-PyTorch turbokv 0.1.0 path. The script's first step is check_model_compatibility(model) — these are the failure modes:

• head_dim must be a power of 2 ∈ {64, 128, 256}. The Randomized Hadamard Transform in fused_turboquant.kernels.triton_rht is built around butterfly operations and has no implementation for arbitrary dims. Verified working on Qwen3 (head_dim=128) and Qwen3.5 text decoder (head_dim=256).
• Separate Q/K/V projections required. Fused-QKV models (qkv_proj, c_attn) are rejected — make_fused_attention_forward raises with a clear error rather than producing garbage.
• No sliding-window attention. Models with sliding_window set on the config or attention module are rejected (e.g., Mistral-Sliding, Gemma-2). The kernel is causal-full-attention only.
• No attention logit softcapping (Gemma-2-style).
• RoPE expected. ALiBi / learned positional embeddings produce incorrect results; check_model_compatibility warns when RoPE isn't detected in config.
• Hybrid linear-attention models (Qwen3.5*, ``): only the full-attention layers (typically 1-in-4) are patched; the linear layers keep their recurrent state. The compatibility checker reports compatible=True because the Triton path will run, but the savings scale only with the full-attention layer count. Note: the bonus Qwen3.5-0.8B run in our local test failed at the baseline generate step (HF DynamicCache is not the right cache for a hybrid model — it raises has_previous_state can only be called on LinearAttention layers); fused-TurboQuant is orthogonal to that issue. Hybrid models need a HybridCache wrapped around CompressedKVCache, which is upstream work. We document the gap and skip the bonus run.

Blackwell / RTX 5080 Laptop notes

The 5080 Laptop is sm_120 (Blackwell), CUDA 13.0, torch 2.11. Triton 3.6.0 supports this architecture. The build does, however, require Python development headers to compile its cuda_utils.so shim on first use:

sudo apt install python3.12-dev

If the headers are missing the kernel raises a confusing error nested inside the smoke test — see the "Last measured run" section below for the exact failure mode. Without the headers the JIT cannot build and patch_model(..., verify=True) fails on the very first forward pass.

Validation

scripts/quantization/test_fused_turboquant.py runs three paths back-to- back on Qwen/Qwen3.5-0.8B with 5 prompts × 128 new tokens at a 4096-token prompt. It writes the full report to scripts/quantization/fused_turboquant_report.json and asserts:

• fused peak VRAM ≤ pure-PyTorch turbokv peak VRAM, and
• fused tok/s ≥ 1.5x pure-PyTorch turbokv tok/s.

uv run python scripts/quantization/test_fused_turboquant.py

Last measured run (Qwen3.5-0.8B, 5080 Laptop, 4-bit, 4096-token prompt + 128 new tokens)

The fused path failed at patch_model(..., verify=True) during the first single-token forward pass. The exact error from /usr/bin/gcc was:

fatal error: Python.h: No such file or directory
    7 | #include <Python.h>
      |          ^~~~~~~~~~

Fix: install the matching Python development headers system-wide so Triton's CUDA utility helper can compile:

sudo apt install python3.12-dev

After installing, re-run the test — the same script will exercise the Triton path without modification. The pure-PyTorch numbers above are real and confirm that turbokv 0.1.0 still delivers the documented memory win (-19.2% peak VRAM vs baseline at 4096-token prompt) but at a 3.48x throughput cost; recovering that throughput is exactly what the Triton kernels in fused-turboquant exist to do.

QJL

Reference: Amir Zandieh, Majid Daliri, Insu Han, QJL: 1-Bit Quantized JL Transform for KV Cache Quantization with Zero Overhead, arXiv:2406.03482, AAAI 2025 (acm dl). Implementation: vendored from amirzandieh/QJL @ 648b3641f96b6e95e091217220b94e4739fd4d82 under scripts/quantization/qjl/ (Apache 2.0 — see scripts/quantization/qjl/LICENSE and qjl/NOTICE.md). There is no pyqjl package on PyPI — the QJL kernel is shipped as a CUDA C++ extension that the user has to compile against their local nvcc. pyproject.toml adds pybind11>=2.12.0 to the train extras so the template-binding step has its dep; nvcc and the matching pythonX.Y-dev headers come from the system package manager.

What this is — and what it is NOT

QJL is the K (keys) side companion to TurboQuant's V (values) side. Both compressors were authored by Amir Zandieh; they are explicitly designed to compose. The combination delivers ~10x KV-cache compression at long context: 1-bit per JL-projected key coordinate + 4-bit per value coordinate, plus a small per-token bf16 norm on each side and an amortized outlier sketch shared across group_size consecutive tokens.

QJL is not a weight quantizer. model.safetensors is unchanged after applying. qjl_apply.py writes the merged base model unchanged plus a qjl_config.json sidecar that records the projection geometry (projection_dim_per_head, projection_seed, outlier_count_*, initial_layers_count, group_size, buffer_size, paired value_bits) so the inference loader can deterministically reconstruct the JL projection matrix Π and the value-side codebook.

Algorithm in one paragraph

For each cached key vector k of length head_dim (per attention head, per token): (1) extract ||k|| and store it in bf16; (2) multiply by a fixed-per-layer Johnson–Lindenstrauss matrix Π ∈ R^{head_dim × s} drawn from N(0, 1) and orthogonalized via a chunked QR decomposition (QJLSketch.init_rot_dir), giving a sketch s = Πᵀ k of length s = projection_dim_per_head; (3) take the sign of each sketch coordinate, packing 8 signs into one uint8. To recover an unbiased estimator of inner products at attention time, the query is projected through the same Π and the score is reconstructed in a custom CUDA kernel (qjl_kernel/csrc/qjl_score_kernel.cu). The paper proves the resulting cosine-similarity estimator has minimal relative distortion at 1 bit. To handle outlier coordinates (a few head_dim indices with disproportionately large norms — common on layer 0 in Llama / Qwen models), the kernel additionally stores a top-k outlier sketch per group of group_size consecutive tokens, with its own larger JL projection of dimension dim_outlier (256 for general layers, 128 for the first initial_layers_count layers). The recent buffer_size tokens stay in fp16 to preserve freshly-generated context losslessly.

Tradeoffs

• Pros. Provably minimal relative distortion at 1-bit (paper Theorem 3). Storage is purely the packed sign sketch plus a per-token bf16 norm — no quantization scale/zero-point per coordinate, which is what gives the "zero overhead" tagline. Hand-written CUDA kernels for both quantize and score paths recover throughput at long context (the paper reports 2.0–2.5x speedup on prefill and decode for Llama-2/3 at 32k context). Composes cleanly with TurboQuant on the V side for end-to-end ~4x KV reduction whole-model.
• Cons. The compression unit is the JL sketch dimension, not the number of bits per coord — at the canonical projection_dim_per_head=256 the K-side ratio head_dim*2 / (projection_dim/8 + 2) works out to 7.53x for head_dim=128 (Qwen3.5-0.8B / Llama-3 / Qwen3.5-2B), not the marketing-headline 16x (which would assume zero norm overhead). Pushing to projection_dim_per_head=128 recovers ~14.2x at the cost of attention-score quality. The kernel hard-codes EMB_DIM 128 in csrc/qjl_quant_kernel.cu (line 7), so other head_dim values need a kernel rebuild with a different define. The CUDA kernel was written for Ampere/Hopper; Blackwell (sm_120) requires the PTX-fallback build flag (see "Build" below).

Supported architectures

The vendored LlamaAttention_QJL and LlamaDecoderLayer_QJL modules under scripts/quantization/qjl/ (mirrored from upstream models/llama3_qjl.py) target the Llama-3 attention shape directly: q_proj / k_proj / v_proj / o_proj linear projections plus LlamaRotaryEmbedding and grouped-query attention (the GQA score kernel cuda_qjl_gqa_score handles num_attention_heads != num_key_value_heads). Verified upstream on:

• Llama-2 7B and Llama-3 8B (the upstream run_longbench.py evaluation set)

Should work, by structural match, on:

• Qwen2 / Qwen2.5 (full-attention, GQA, head_dim=128)
• Qwen3 0.8B / 2B / 4B / 8B (full-attention, GQA, head_dim=128) — validated locally by the pure-PyTorch ratio probe in test_qjl.py

Qwen3.5 caveat (read this)

The active targets Qwen3.5-{0.8B, 2B, 4B} are **hybrid linear-attention

• Gated Attention** models. QJL only applies to full_attention layers — there is nothing to compress in a recurrent state. The qjl_apply.py calibration step honors layer_types and silently skips linear-attention layers. The on-disk config records n_full_attention_layers so the inference loader knows which layers to wrap.

The 0.8B / 2B / 4B vision-language variants (Qwen3_5ForConditionalGeneration) need to be loaded with AutoModelForVision2Seq and the text decoder extracted via model.language_model before patching the attention modules. We have not done that integration yet — test_qjl.py falls back to Qwen/Qwen3.5-0.8B for the 0.8B-class validation.

Build

The QJL kernel is not pip-installable. Build it once, in place, inside the vendored directory:

cd scripts/quantization/qjl
# Default Ampere/Hopper build:
python setup.py build_ext --inplace
# Blackwell (RTX 50-series, sm_120) — the kernel was not authored
# against this arch; force PTX fallback so it compiles for compute_120
# via the JIT path:
TORCH_CUDA_ARCH_LIST="12.0+PTX" python setup.py build_ext --inplace

Prerequisites (verified missing on the local 5080 dev box at the time this was vendored):

sudo apt install nvidia-cuda-toolkit python3.12-dev

The first installs nvcc (PyTorch wheels ship ptxas and nvrtc but not the full nvcc driver — the cu13 wheel directory only contains include/, lib/, no bin/nvcc). The second installs Python.h so the C++ extension's pybind11 bindings can compile.

CLI

Apply to a fine-tuned 2B checkpoint (auto-merges if --model points to a LoRA adapter):

uv run python scripts/quantization/qjl_apply.py \
    --model checkpoints/qwen35-2b-eliza/final \
    --calibration data/final/val.jsonl \
    --calibration-samples 128 \
    --output checkpoints/qwen35-2b-eliza/final-qjl

Apply to a fine-tuned 27B checkpoint (same shape; calibration is single-pass forward and fits in 16 GB only with offload):

uv run python scripts/quantization/qjl_apply.py \
    --model checkpoints/qwen35-4b-eliza/final \
    --calibration data/final/val.jsonl \
    --calibration-samples 128 \
    --projection-dim-per-head 256 \
    --projection-dim-per-head-initial 512 \
    --initial-layers-count 15 \
    --outlier-count-general 8 \
    --value-bits 4 \
    --output checkpoints/qwen35-4b-eliza/final-qjl

Useful knobs:

• --key-bits 1 (only). QJL is a 1-bit-per-projected-coord scheme by construction; the flag exists for forward-compat.
• --projection-dim-per-head N (default 256). The JL sketch dimension per attention head. Smaller = more compression, lower attention score fidelity. Must be byte-aligned (% 8 == 0).
• --projection-dim-per-head-initial N (default 512). The first --initial-layers-count layers carry more attention mass; the paper recommends a larger sketch budget there. Leaving this at 2× the general dim matches upstream defaults exactly.
• --initial-layers-count N (default 15, paper's choice for Llama-7B with 32 layers). For different layer counts, scale ~half.
• --outlier-count-general K / --outlier-count-initial-layers K (default 8 each). Top-K head_dim coords promoted to the outlier sketch per group. The calibration step measures per-layer outlier norm ratio and records it in the sidecar so the inference loader can validate this budget.
• --value-bits {2,4} (default 4). Companion TurboQuant V-side bits recorded in the sidecar.
• --group-size N (default 32). Group of consecutive tokens that share an outlier-coord index table. Larger = smaller per-token outlier overhead, slightly worse outlier tracking on bursty activations.
• --projection-seed N (default 42). PRNG seed for the JL matrix. Pin this to the value recorded in qjl_config.json at inference time; otherwise Π won't match and the dequantized cosine scores are garbage.

Validation

scripts/quantization/test_qjl.py runs on Qwen/Qwen3.5-0.8B (closest text-only stand-in for Qwen/Qwen3.5-0.8B — see caveat above):

1. Attempts to build the vendored CUDA extension. If nvcc or Python.h is missing it records the exact remediation command (sudo apt install nvidia-cuda-toolkit python3.12-dev) and skips the runtime-kernel path.
2. Runs a baseline bf16 generation and records peak VRAM, tok/s, and sample outputs.
3. Captures the K activations from the first 4 attention layers via a forward hook, runs a pure-PyTorch reference QJL quantize on them (no CUDA extension needed — JL @ matmul + sign + bit-packing in plain PyTorch), and reports the realized K-side compression ratio per layer.
4. Sweeps projection_dim ∈ {128, 256, 512} so the report shows the full size-vs-quality curve at the same K activations.
5. Computes the analytic whole-model KV-bytes-per-token reduction with the paired V-side TurboQuant numbers via qjl_apply.kv_bytes_per_token_analytic.
6. Asserts (a) K-side ratio ≥ 7× at the canonical 256-dim setting, matching the closed-form head_dim*2 / (projection_dim/8 + 2) = 7.53× for head_dim=128, and (b) baseline outputs are non-degenerate.

uv run python scripts/quantization/test_qjl.py

The full numeric report is written to scripts/quantization/qjl_report.json. Last measured run on Qwen3.5-0.8B / 5080 (bf16 baseline cache, projection_dim=256, seed=42):

Blockers (current state on the 5080 dev box)

• nvcc is not on the system. PyTorch's bundled cu13 wheel ships nvrtc and ptxas but no full nvcc driver, and the nvidia-cuda-nvcc-cu12 PyPI wheel only ships ptxas (verified). Fix: sudo apt install nvidia-cuda-toolkit.
• Python.h is not present (no python3.12-dev package installed). Fix: sudo apt install python3.12-dev.
• Blackwell (sm_120) is not in the upstream test matrix. The kernel sources are written against Ampere/Hopper. After the two apt installs above, the recommended build command is TORCH_CUDA_ARCH_LIST="12.0+PTX" python setup.py build_ext --inplace, which forces the JIT-PTX fallback path; the kernel will compile to compute_120 via PTX at first load. If the kernel still fails at runtime on sm_120, the immediate workaround is to run the pure-PyTorch reference path in test_qjl.py:qjl_pure_pytorch_quantize for measurement and ship the validated checkpoint to a Hopper/Ampere host for actual inference.
• Hard-coded EMB_DIM 128 in qjl/csrc/qjl_quant_kernel.cu:7. The code only works for head_dim == 128 out of the box. Qwen3.5-0.8B / Qwen3.5-2B / Llama-3 all match. If we later need to apply QJL to a model with head_dim != 128, the #define EMB_DIM must be edited and the kernel rebuilt; there is no runtime arg for it.

Abliteration

Reference: Arditi et al., Refusal in LLMs is mediated by a single direction (arXiv:2406.11717). Practical writeup: Maxime Labonne, "Uncensor any LLM with abliteration".

Computes the rank-1 refusal direction r = normalize(mean(harmful) - mean(harmless)) from residual-stream activations at a configurable mid-stack layer, then projects r out of every block's self_attn.o_proj and mlp.down_proj weights so the model can never write the refusal direction back into the residual stream. Destructive transform — save to a NEW directory.

uv run python scripts/quantization/abliteration_apply.py \
    --checkpoint Qwen/Qwen3.5-0.8B \
    --output checkpoints/qwen3.5-0.8b-abliterated \
    --harmful-jsonl data/harmful.jsonl \
    --harmless-jsonl data/harmless.jsonl

Without --harmful-jsonl / --harmless-jsonl the script falls back to a small built-in pair list intended for smoke testing only — pass real prompt corpora (e.g. mlabonne/harmful_behaviors) in production.