Upload README.md

README.md

@@ -10,6 +10,7 @@ tags:
   - gguf
   - llama.cpp
   - bfloat16
+  - q8_0
   - mtp
   - speculative-decoding
   - veriloop
@@ -30,7 +31,7 @@ tags:
 
 <p align="center">
   <strong>Official llama.cpp distribution of VeriLoop E2</strong><br>
-  <em>Reference-grade BF16 target + MTP speculative-decoding draft</em><br><br>
+  <em>Canonical BF16 reference · High-fidelity Q8_0 · evidence-driven quantization releases</em><br><br>
   <strong>27B post-trained model for code, mathematics, and physics · 262K native context · Apache License 2.0</strong><br>
   <strong>Developed by Tsinghua SIGS Robot Lab · Libo Wang</strong>
 </p>
@@ -39,8 +40,9 @@ tags:
   <a href="https://www.apache.org/licenses/LICENSE-2.0"><img src="https://img.shields.io/badge/License-Apache--2.0-2F80ED?style=flat-square" alt="License: Apache 2.0"></a>
   <img src="https://img.shields.io/badge/Format-GGUF-111827?style=flat-square" alt="Format: GGUF">
   <img src="https://img.shields.io/badge/Reference-BF16-5B5BD6?style=flat-square" alt="Reference: BF16">
+  <img src="https://img.shields.io/badge/High--Fidelity-Q8__0-7C3AED?style=flat-square" alt="High Fidelity: Q8_0">
+  <img src="https://img.shields.io/badge/Q8%20Quality%20Gate-PASS-16A34A?style=flat-square" alt="Q8 quality gate: PASS">
   <img src="https://img.shields.io/badge/Runtime-llama.cpp-0A7F6F?style=flat-square" alt="Runtime: llama.cpp">
-  <img src="https://img.shields.io/badge/MTP-Supported-7C3AED?style=flat-square" alt="MTP: Supported">
 </p>
 
 <p align="center">
@@ -56,36 +58,145 @@ tags:
 
 **VeriLoop E2 GGUF** is the official llama.cpp-oriented distribution of **VeriLoop E2**, a 27B post-trained model built on **Qwen3.8-27B** for code, software engineering, mathematics, and physics.
 
-This repository is intentionally organized as a **GGUF release repository rather than a duplicate of the parent model card**. The parent repository remains the authoritative source for model methodology, training, system architecture, benchmark protocols, scientific demonstrations, and release boundaries. This repository focuses on the artifacts and information that matter for local GGUF deployment:
+This repository follows a **reference-first, measurement-driven** release policy:
 
-- canonical GGUF files and their roles;
-- exact conversion provenance and integrity checks;
-- llama.cpp loading and serving instructions;
-- MTP speculative decoding;
-- hardware and context guidance;
-- quantization-quality policy for lower-bit releases;
-- a concise, evidence-linked snapshot of the parent model's public benchmark record.
+- **BF16** is the canonical GGUF reference used for reproducibility, conversion baselines, and quantization-quality measurement.
+- **Q8_0** is the first measured low-bit release and is positioned as the **high-fidelity local sweet spot** for users who want substantially lower storage and memory pressure while remaining very close to the BF16 reference.
+- Future Q6/Q5/Q4 releases will be admitted only after the same BF16-relative fidelity protocol is completed.
 
-The first release establishes **BF16 as the canonical GGUF reference point**. Lower-bit variants are published only after they are compared against this reference under a fixed quality protocol.
+The Q8_0 headline result is:
+
+> **46.86% smaller than BF16, with only +0.0643% relative perplexity increase under the frozen paired protocol; Mean KLD = 0.002176 and Same top-p = 98.815%.**
+
+These are **quantization-fidelity measurements**, not a claim that every downstream benchmark score changes by exactly 0.0643%.
+
+---
+
+## Choose the Right Precision
+
+| Variant | Positioning | Main model size | Fidelity / efficiency role | Recommended for |
+|---|---|---:|---|---|
+| **BF16** | **Reference Fidelity Sweet Spot** | **53.808 GB / 50.113 GiB** | Canonical GGUF reference; no low-bit quantization loss | Research, reproducibility, quantization baselines, high-memory deployment |
+| **Q8_0** | **High-Fidelity Local Sweet Spot** | **28.596 GB / 26.632 GiB** | **46.86% smaller**; measured BF16-relative drift remains very small | Quality-first local inference, workstation/server deployment |
+| `Q6_K` | Quality / efficiency tier | Pending validation | Planned BF16-relative measurement | Users wanting lower memory with very high quality |
+| `Q5_K_M` | Balanced tier | Pending validation | Planned BF16-relative measurement | Quality / memory balance |
+| `Q4_K_M` | Mainstream local tier | Pending validation | Planned BF16-relative measurement | Broad consumer-hardware deployment |
+
+### BF16: reference-fidelity sweet spot
+
+Choose **BF16** when your priority is **reference fidelity, reproducibility, attribution, or quantization research** rather than storage efficiency. It is the canonical GGUF baseline against which every lower-bit VeriLoop E2 release is measured.
+
+BF16 is therefore the repository's **reference sweet spot**, not its memory-efficiency sweet spot.
+
+### Q8_0: high-fidelity local sweet spot
+
+Choose **Q8_0** when you want the closest practical local GGUF experience to BF16 while materially reducing model-file size. Q8_0 cuts the main GGUF from **50.113 GiB to 26.632 GiB** and passes the frozen VeriLoop quantization-quality gates.
+
+Q8_0 is the current **quality-first local recommendation**. It is not presented as the smallest or most memory-efficient option.
+
+---
+
+## Available Files
+
+### Main models
+
+| File | Precision / quant | Exact size | Role | Status |
+|---|---|---:|---|---|
+| `VeriLoop-E2-BF16.gguf` | BF16 / selected F32 tensors | **53.808 GB / 50.113 GiB** | Canonical reference | **Verified** |
+| `VeriLoop-E2-Q8_0.gguf` | Q8_0 / retained F32 tensors | **28.596 GB / 26.632 GiB** | High-fidelity local quant | **Verified / Quality Gate PASS** |
+
+### Optional MTP draft files
+
+| File | Role | Approx. size | Notes |
+|---|---|---:|---|
+| `mtp-VeriLoop-E2-BF16.gguf` | BF16 MTP companion | ~6.04 GB | Optional speculative-decoding artifact |
+| `mtp-VeriLoop-E2-Q8_0.gguf` | Q8 release MTP companion | **3.287 GB / 3.061 GiB** | Stock llama.cpp load + speculative-draft smoke validated |
+
+Q8 release identity, exact hashes, and byte counts are recorded in [`Q8_RELEASE_MANIFEST.json`](./Q8_RELEASE_MANIFEST.json).
 
 ---
 
-## Release Files
+## Q8_0 Performance-Loss Benchmark — BF16 Reference vs Q8_0
+
+The Q8_0 release is evaluated **directly against the canonical BF16 GGUF** under the same llama.cpp comparison protocol.
 
-| File | Role | Precision | Tensor count | Approx. local size | SHA256 | Status |
-|---|---|---:|---:|---:|---|---|
-| `VeriLoop-E2-BF16.gguf` | Canonical target model | BF16 / selected F32 metadata-sensitive tensors | **851** | **~51 GiB** | `11bf5defde1a256b7582bc34fd2c4a85a61615ed88e7a422dcd24d814ea6d35d` | **Verified** |
-| `mtp-VeriLoop-E2-BF16.gguf` | MTP draft for speculative decoding | BF16 | **18** | **~5.6 GiB** | `10e36ca18d508f3321f976262cc2e05b01e807f3a797605696a8b6a3f490429f` | **Verified** |
+### Headline fidelity
 
-> **BF16 is the reference, not a low-bit quantization.** It is provided as the high-fidelity GGUF baseline against which subsequent Q8/Q6/Q5/Q4/IQ4 releases can be measured.
+| Metric | BF16 reference | Q8_0 | Q8_0 drift |
+|---|---:|---:|---:|
+| Mean PPL | **4.840423 ± 0.119931** | **4.843536 ± 0.120062** | **+0.003113 / +0.0643%** |
+| PPL ratio | 1.000000 | **1.000643 ± 0.000754** | **+0.0643%** |
+| Mean KLD vs BF16 | 0 | **0.002176 ± 0.000668** | lower is better |
+| Same top-p | 100% reference | **98.815 ± 0.120%** | **1.185 pp disagreement** |
+| RMS token-probability Δp | 0% reference | **1.251 ± 0.137%** | distributional-noise scale |
+| Mean token-probability Δp | 0% reference | **−0.007 ± 0.014%** | near-zero directional bias |
+| log-PPL correlation | 100% reference | **99.95%** | very high agreement |
 
-Hugging Face may display file sizes in decimal GB while local Unix tools commonly display binary GiB; small differences in displayed size are therefore expected.
+> **Release interpretation:** Q8_0 removes **46.86%** of the BF16 GGUF file footprint while the paired PPL ratio remains **1.000643**, Mean KLD is **0.002176**, and Same top-p agreement is **98.815%**. For this quantization tier, these BF16-paired distribution metrics are the primary evidence of quantization fidelity.
+
+### Release gates
+
+| Gate | Threshold | Measured | Result |
+|---|---:|---:|---|
+| PPL ratio | ≤ 1.005 | **1.000643** | **PASS** |
+| Mean KLD | ≤ 0.005 | **0.002176** | **PASS** |
+| Same top-p | ≥ 97.0% | **98.815%** | **PASS** |
+| Main tensor audit | 851 tensors | **851** | **PASS** |
+| llama.cpp load + generation | required | **PASS** | **PASS** |
+| Q8 MTP load + speculative draft path | required | **PASS** | **PASS** |
+
+The numerical thresholds above are **VeriLoop release-QA gates**, not universal industry standards.
+
+For the complete measurement record, see **[`QUANTIZATION_QUALITY.md`](./QUANTIZATION_QUALITY.md)**.
 
 ---
 
-## Conversion Provenance and Quality Gates
+## Fixed Q8 Comparison Protocol
 
-This release was converted with the public **llama.cpp** GGUF toolchain and validated before promotion.
+| Item | Value |
+|---|---|
+| Baseline | Canonical VeriLoop E2 BF16 GGUF |
+| Candidate | VeriLoop E2 Q8_0 GGUF |
+| Corpus | WikiText-2 raw test |
+| Context | 2,048 tokens |
+| Chunks | 8 |
+| Seed | 42 |
+| GPU layers | 40 |
+| Flash Attention | Off |
+| KV cache | F16 / F16 |
+| Batch / micro-batch | 512 / 512 |
+| Evaluation tool | `llama-perplexity` |
+| Q8 quantization / QC llama.cpp revision | `42916d83f4a225e56709f873aa8050ac11f5b6a4` |
+| BF16 reference SHA256 | `11bf5defde1a256b7582bc34fd2c4a85a61615ed88e7a422dcd24d814ea6d35d` |
+
+The BF16 logits were recorded once and used as the fixed reference for Q8_0 KL-divergence evaluation.
+
+---
+
+## Q8_0 Quantization Configuration
+
+| Property | Value |
+|---|---|
+| Source | Canonical VeriLoop E2 BF16 GGUF |
+| Quantization | `Q8_0` |
+| Source model size reported by quantizer | 51,305.09 MiB |
+| Quantized model size reported by quantizer | 27,260.56 MiB |
+| Effective density reported by quantizer | **8.50 BPW** |
+| Main tensor count | **851** |
+| Q8_0 tensors | **498** |
+| Retained F32 tensors | **353** |
+| Quantization threads | 25 |
+| Quantization time | **38.315 s** |
+| Quantization return code | **0** |
+| Main build gate | **PASS** |
+
+The retained F32 tensors are part of llama.cpp's quantized output layout; `Q8_0` does not imply that every tensor in the file is stored in the same integer representation.
+
+---
+
+## Conversion and Release Provenance
+
+### BF16 reference
 
 | Check | Result |
 |---|---|
@@ -94,53 +205,30 @@ This release was converted with the public **llama.cpp** GGUF toolchain and vali
 | HF architecture | `Qwen3_5ForConditionalGeneration` |
 | Native model context | 262,144 tokens |
 | Conversion tool | `convert_hf_to_gguf.py` |
-| llama.cpp source revision | `bddf8263c31c3dce3212263b00ebd2d98c1a752b` |
-| Target export | BF16, separate from MTP |
-| MTP export | BF16 draft GGUF |
-| Target conversion return code | **0** |
-| MTP conversion return code | **0** |
-| Target tensor audit | **851 tensors / PASS** |
-| MTP tensor audit | **18 tensors / PASS** |
+| BF16 conversion llama.cpp revision | `bddf8263c31c3dce3212263b00ebd2d98c1a752b` |
+| BF16 target tensor audit | **851 tensors / PASS** |
 | Non-zero tensor audit | **PASS** |
 | Source-checkpoint immutability gate | **PASS** |
-| CUDA llama-server build | **PASS** |
-| llama-server model load | **PASS** |
-| OpenAI-compatible `/v1/models` endpoint | **PASS** |
-| Reference GGUF runtime context tested in release gate | **32,768 tokens** |
-
-The native 262K context length is inherited from the parent model configuration. The **32K figure above is the context length used in this GGUF release's runtime validation**, not a redefinition of the model's native context. Larger llama.cpp contexts require appropriate VRAM/RAM and KV-cache planning and should be validated on the target hardware.
+| BF16 main-model load gate | **PASS** |
 
-### Integrity policy
+### Q8_0 release
 
-The release process treats conversion as an auditable transformation rather than a successful file write:
-
-```text
-Parent checkpoint
-    ↓
-tensor-name / shard consistency checks
-    ↓
-BF16 target export
-    ↓
-separate MTP export
-    ↓
-GGUF tensor-count audit
-    ↓
-llama-server build + load gate
-    ↓
-runtime endpoint gate
-    ↓
-release manifest + SHA256
-```
-
-The SHA256 values above are the frozen identities for the initial BF16 reference release.
+| Check | Result |
+|---|---|
+| Quantization source | Canonical BF16 GGUF |
+| Requantization from Q8 / lower precision | **No** |
+| Quantization type | **Q8_0** |
+| llama.cpp quantization / QC revision | `42916d83f4a225e56709f873aa8050ac11f5b6a4` |
+| Structure gate | **PASS** |
+| Quantitative fidelity gate | **PASS** |
+| Main-model runtime generation | **PASS** |
+| Q8 MTP compatibility smoke | **PASS** |
 
 ---
 
 ## Parent-Model Benchmark Record
 
-The following numbers are the **frozen public VeriLoop E2 parent-model release results**. They are included here so local users can identify the model they are downloading without having to reconstruct its evaluation record from multiple pages.
-
-**These scores are not presented as GGUF-specific re-runs.** BF16 is a format-conversion reference; each future lower-bit variant will be evaluated for degradation relative to this BF16 reference before being recommended.
+The following figure and scores identify the **parent VeriLoop E2 release**. They are not relabeled as Q8_0-specific task scores; Q8_0 quantization fidelity is measured directly against BF16 above.
 
 <p align="center">
   <img src="https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2/resolve/main/.eval_results/veriloop_e2_benchmark_result.png" width="100%" alt="VeriLoop E2 benchmark comparison across nine public benchmarks">
@@ -153,7 +241,7 @@ The following numbers are the **frozen public VeriLoop E2 parent-model release r
 | DeepSWE v1.1 | **64.6** | [Evidence](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/deepswe-1.1) |
 | Terminal-Bench 3.0 | **29.7** | [Evidence](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/terminal-bench-3.0) |
 | Terminal-Bench 4.0 | **37.9** | [Evidence](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/terminal-bench-4.0) |
-| SWE-Marathon v1.1 | **45.0** | [Release source](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2) |
+| SWE-Marathon v1.1 | **45.0** | [Parent release](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2) |
 | AIME 2026 | **98.3** | [Evidence](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/aime-2026) |
 | GPQA Diamond | **93.9** | [Evidence](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/gpqa-diamond) |
 | Apex 2025 | **89.6** | [Evidence](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence/tree/main/apex-2025) |
@@ -181,116 +269,96 @@ cmake -B build -DGGML_CUDA=ON
 cmake --build build -j --target llama-server llama-cli
 ```
 
-### Download the BF16 reference
+### Download BF16
 
 ```bash
-hf download \
-  tsinghua-sigs-robot-lab/VeriLoop-E2-GGUF \
-  VeriLoop-E2-BF16.gguf \
-  --local-dir .
+hf download   tsinghua-sigs-robot-lab/VeriLoop-E2-GGUF   VeriLoop-E2-BF16.gguf   --local-dir .
 ```
 
-### Run the target model
+### Download Q8_0
 
 ```bash
-llama-server \
-  -m ./VeriLoop-E2-BF16.gguf \
-  -ngl all \
-  -c 32768 \
-  --host 127.0.0.1 \
-  --port 8080
+hf download   tsinghua-sigs-robot-lab/VeriLoop-E2-GGUF   VeriLoop-E2-Q8_0.gguf   --local-dir .
 ```
 
-The server exposes an OpenAI-compatible API at:
+### Run BF16
 
-```text
-http://127.0.0.1:8080/v1
+```bash
+llama-server   -m ./VeriLoop-E2-BF16.gguf   -ngl all   -c 32768   --host 127.0.0.1   --port 8080
 ```
 
-### Direct CLI inference
+### Run Q8_0
 
 ```bash
-llama-cli \
-  -m ./VeriLoop-E2-BF16.gguf \
-  -ngl all \
-  -c 32768
+llama-server   -m ./VeriLoop-E2-Q8_0.gguf   -ngl all   -c 32768   --host 127.0.0.1   --port 8080
 ```
 
-`-ngl all` is appropriate when the full model fits in available VRAM. llama.cpp also supports automatic or partial GPU offload; users with less VRAM should allow the runtime to place part of the model in system memory.
+The OpenAI-compatible endpoint is available at:
+
+```text
+http://127.0.0.1:8080/v1
+```
+
+`-ngl all` is appropriate only when the model, KV cache, and runtime workspace fit the available accelerator memory. Partial GPU offload is supported when full offload is not practical.
 
 ---
 
-## MTP Speculative Decoding
+## Optional Q8_0 MTP Speculative Decoding
 
-The release includes a separate BF16 MTP draft GGUF. In llama.cpp, MTP can be used as a speculative draft model:
+Download the validated Q8_0 MTP companion:
 
 ```bash
-hf download \
-  tsinghua-sigs-robot-lab/VeriLoop-E2-GGUF \
-  mtp-VeriLoop-E2-BF16.gguf \
-  --local-dir .
+hf download   tsinghua-sigs-robot-lab/VeriLoop-E2-GGUF   mtp-VeriLoop-E2-Q8_0.gguf   --local-dir .
 ```
 
+Validated resource-conservative smoke configuration:
+
 ```bash
-llama-server \
-  -m ./VeriLoop-E2-BF16.gguf \
-  --model-draft ./mtp-VeriLoop-E2-BF16.gguf \
-  --spec-type draft-mtp \
-  -ngl all \
-  --n-gpu-layers-draft all \
-  -c 32768 \
-  --host 127.0.0.1 \
-  --port 8080
+llama-server   -m ./VeriLoop-E2-Q8_0.gguf   --model-draft ./mtp-VeriLoop-E2-Q8_0.gguf   --spec-type draft-mtp   -ngl all   --n-gpu-layers-draft 0   -c 8192   --host 127.0.0.1   --port 8080
 ```
 
-MTP is an **optional acceleration path**. Throughput gains depend on hardware, prompt distribution, draft acceptance, context length, and llama.cpp version; this repository does not claim a universal speedup factor.
+MTP is an optional acceleration path. Throughput gains depend on hardware, prompt distribution, draft acceptance, context length, and llama.cpp revision; no universal speedup factor is claimed.
 
 ---
 
 ## Memory Guidance
 
-The BF16 reference is intentionally large.
-
-| Component | Approx. file size | Practical implication |
+| Variant | Model file | Practical implication |
 |---|---:|---|
-| Main BF16 GGUF | ~51 GiB | Requires substantial GPU VRAM or CPU RAM |
-| BF16 MTP draft | ~5.6 GiB | Additional memory when speculative decoding is enabled |
+| BF16 main | **50.113 GiB** | Reference-quality deployment; substantial VRAM/RAM required |
+| Q8_0 main | **26.632 GiB** | Much lower model footprint while retaining measured near-reference fidelity |
+| Q8_0 MTP companion | **3.061 GiB** | Additional memory only when speculative decoding is enabled |
 | KV cache / runtime workspace | Workload-dependent | Increases with context length, batch size, cache dtype, and parallelism |
 
-A full-GPU BF16 deployment therefore needs materially more memory than the model file alone. For constrained systems, wait for or use a validated lower-bit release rather than forcing an unsuitable BF16 configuration.
+Model-file size is not total runtime-memory consumption. Leave headroom for KV cache, compute buffers, runtime metadata, and optional draft-model execution.
 
 ---
 
 ## Quantization Release Policy
 
-This repository follows a **reference-first, measurement-driven** quantization policy.
+Every lower-bit VeriLoop E2 release is compared against the **same canonical BF16 reference**. A quant is not labeled “recommended” solely because the format is common.
 
-The BF16 GGUF above is the baseline. Lower-bit releases are not labeled as recommended solely because they are common quant types. Each candidate is evaluated against the BF16 reference before recommendation.
+Planned product ladder:
 
-The release program targets the following families:
+| Quant | Intended role | Publication rule |
+|---|---|---|
+| **Q8_0** | High-fidelity local sweet spot | **Released — quantitative gate PASS** |
+| `Q6_K` | Quality / efficiency tier | Publish only after BF16-paired QC |
+| `Q5_K_M` | Balanced tier | Publish only after BF16-paired QC |
+| `Q4_K_M` | Mainstream local tier | Publish only after BF16-paired QC |
+| `IQ4_XS` | Compact 4-bit-class tier | Publish only after BF16-paired QC |
 
-- **Q8_0** — high-fidelity local deployment;
-- **Q6_K** — quality-focused compression;
-- **Q5_K_M / Q5_K_S** — high-quality balanced tiers;
-- **Q4_K_M / Q4_K_S** — broad local-deployment tiers;
-- **IQ4_NL / IQ4_XS** — compact 4-bit-class alternatives where supported.
+The comparison protocol prioritizes:
 
-For each lower-bit release, the quality card is intended to report:
+- file size / effective BPW;
+- PPL ratio vs BF16;
+- Mean KL divergence vs BF16;
+- token-probability drift;
+- Same top-p agreement;
+- structural integrity;
+- llama.cpp load and generation compatibility.
 
-| Dimension | Measurement |
-|---|---|
-| File size | GB / GiB |
-| Effective density | bits per weight where applicable |
-| Language-model drift | fixed-corpus perplexity delta vs BF16 |
-| Distribution drift | mean KL divergence vs BF16 |
-| Decision stability | same-top-p / token agreement metrics |
-| Code capability | selected software-engineering regression gates |
-| Mathematics | selected mathematical-reasoning regression gates |
-| Science | selected scientific-reasoning regression gates |
-| Long-context behavior | fixed-context smoke / stability checks |
-| llama.cpp runtime | load, generation, API and MTP compatibility |
-
-**No cross-repository PPL or KLD comparison is claimed unless the corpus, tokenizer path, chunking, context, and command line are identical.** Absolute perplexity numbers from different GGUF repositories are generally not directly comparable.
+**No cross-repository PPL or KLD comparison is claimed unless the corpus, tokenizer path, chunking, context, and runtime protocol are identical.**
 
 ---
 
@@ -309,29 +377,28 @@ For each lower-bit release, the quality card is intended to report:
 | Native context | 262,144 tokens |
 | Languages | English, Chinese |
 | GGUF runtime | llama.cpp |
-| Initial GGUF reference | BF16 |
-| Speculative decoding | MTP draft supplied separately |
+| Reference precision | BF16 |
+| Released quantization | Q8_0 |
 | License | Apache License 2.0 |
 
 ---
 
 ## Prompt and Chat Protocol
 
-Use the chat template embedded in the GGUF / inherited from the VeriLoop E2 release. Do not casually replace role delimiters, reasoning delimiters, stop conditions, or tool-call formatting: prompt-template changes can materially alter observed behavior even when the model weights are identical.
+Use the chat template embedded in the GGUF / inherited from VeriLoop E2. Prompt-template changes, reasoning delimiters, stop rules, samplers, cache settings, and tool-call parsers can materially change observed behavior even when model weights are identical.
 
 For the full model protocol and system-level behavior, see the **[VeriLoop E2 parent model card](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2)**.
 
 ---
 
-## Release Boundaries
+## Measurement Boundaries
 
-- This repository distributes the **GGUF deployment form** of VeriLoop E2.
-- The BF16 file is a reference conversion, not a claim of low-bit compression.
-- Parent-model benchmark numbers above are provided for model identity and provenance; lower-bit GGUFs require their own degradation measurements.
-- The model may still produce incorrect code, invalid mathematical reasoning, unsupported scientific arguments, or unsafe commands.
-- Native context length does not guarantee that every local hardware configuration can serve the full context efficiently.
+- **+0.0643% is the measured relative Q8_0 PPL increase under the fixed paired protocol; it is not a universal capability-loss percentage.**
+- Parent-model leaderboard results are included for model identity and provenance, not represented as Q8_0 task re-runs.
+- PPL/KLD values are meaningful for the paired BF16→Q8_0 comparison because both variants use the same evaluation setup.
+- The native 262K context comes from the parent model configuration. Practical local context depends on available memory and runtime settings.
 - MTP acceleration is runtime- and workload-dependent.
-- Community-modified templates, samplers, stop rules, cache settings, or parsers can materially change observed behavior.
+- The model can still produce incorrect code, mathematical reasoning, scientific arguments, or commands.
 
 ---
 
@@ -341,6 +408,8 @@ For the full model protocol and system-level behavior, see the **[VeriLoop E2 pa
 |---|---|
 | Parent model | [VeriLoop E2](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2) |
 | GGUF repository | [VeriLoop E2 GGUF](https://huggingface.co/tsinghua-sigs-robot-lab/VeriLoop-E2-GGUF) |
+| Q8 quality record | [Quantization Quality](./QUANTIZATION_QUALITY.md) |
+| Q8 release identity | [Q8 Release Manifest](./Q8_RELEASE_MANIFEST.json) |
 | Technical report | [OpenReview](https://openreview.net/forum?id=P6FIQILHwX&noteId=P6FIQILHwX) |
 | Evaluation evidence | [VeriLoop E2 Evaluation Evidence](https://huggingface.co/datasets/tsinghua-sigs-robot-lab/VeriLoop-E2-Evaluation-Evidence) |
 | Riemann ζ artifact | [Public artifact](https://github.com/brucewang123456789/GeniusTrail/tree/VeriLoop-E2/riemann-hypothesis) |
@@ -371,7 +440,7 @@ If you use VeriLoop E2 or this GGUF distribution in research, please cite the pa
 }
 ```
 
-For quantization-specific comparisons, identify the exact GGUF filename and SHA256 in addition to the parent-model citation.
+For quantization-specific comparisons, identify the exact GGUF filename and release manifest.
 
 ---
 
@@ -379,4 +448,4 @@ For quantization-specific comparisons, identify the exact GGUF filename and SHA2
 
 VeriLoop E2 builds on **Qwen3.8-27B** and the open-source model-serving and evaluation ecosystem. We thank the communities behind Qwen, llama.cpp, Transformers, Hugging Face, software-engineering benchmarks, mathematical evaluation suites, and reproducible scientific computing.
 
-This GGUF repository is maintained as a **reproducible deployment release**, with conversion provenance, immutable file identities, and quantization-quality measurements treated as first-class release artifacts.
+This repository treats **reference fidelity, quantization quality, immutable artifact identity, and reproducible runtime validation as first-class release properties**.