mirror of
https://github.com/onestardao/WFGY.git
synced 2026-04-28 11:40:07 +00:00
Create exllamaV2.md
This commit is contained in:
PSBigBig
GitHub
parent
0bb1c4ac04
commit
3bc08d3b63
1 changed files with 131 additions and 0 deletions
131
ProblemMap/GlobalFixMap/LocalDeploy_Inference/exllamaV2.md
Normal file
131
ProblemMap/GlobalFixMap/LocalDeploy_Inference/exllamaV2.md
Normal file
|
|
@ -0,0 +1,131 @@
|
|||
# ExLlamaV2: Guardrails and Fix Patterns
|
||||
|
||||
ExLlamaV2 is a specialized inference backend for LLaMA-family models with optimized 4-bit quantization.
|
||||
It provides faster throughput and lower VRAM usage compared to generic backends, but introduces new risks in accuracy, schema drift, and numerical stability.
|
||||
This page maps those issues to WFGY structural fixes with measurable acceptance targets.
|
||||
|
||||
---
|
||||
|
||||
## Open these first
|
||||
- Visual map and recovery: [RAG Architecture & Recovery](https://github.com/onestardao/WFGY/blob/main/ProblemMap/rag-architecture-and-recovery.md)
|
||||
- End-to-end retrieval knobs: [Retrieval Playbook](https://github.com/onestardao/WFGY/blob/main/ProblemMap/retrieval-playbook.md)
|
||||
- Embedding vs meaning: [Embedding ≠ Semantic](https://github.com/onestardao/WFGY/blob/main/ProblemMap/embedding-vs-semantic.md)
|
||||
- Chunk schema: [Chunking Checklist](https://github.com/onestardao/WFGY/blob/main/ProblemMap/chunking-checklist.md)
|
||||
- Collapse and entropy: [Logic Collapse](https://github.com/onestardao/WFGY/blob/main/ProblemMap/logic-collapse.md), [Entropy Collapse](https://github.com/onestardao/WFGY/blob/main/ProblemMap/entropy-collapse.md)
|
||||
- Ordering and boot issues: [Bootstrap Ordering](https://github.com/onestardao/WFGY/blob/main/ProblemMap/bootstrap-ordering.md), [Pre-deploy Collapse](https://github.com/onestardao/WFGY/blob/main/ProblemMap/predeploy-collapse.md)
|
||||
|
||||
---
|
||||
|
||||
## Core acceptance
|
||||
- ΔS drift vs FP16 baseline ≤ 0.10
|
||||
- Coverage ≥ 0.70 for target section
|
||||
- λ convergent across 3 paraphrases and 2 seeds
|
||||
- Latency improvement ≥ 25% with accuracy loss ≤ 5%
|
||||
|
||||
---
|
||||
|
||||
## Typical ExLlamaV2 breakpoints → exact fix
|
||||
|
||||
| Symptom | Likely cause | Open this |
|
||||
|---|---|---|
|
||||
| Text fluency high, citations missing | Schema loosened in quantized path | [Data Contracts](https://github.com/onestardao/WFGY/blob/main/ProblemMap/data-contracts.md), [Retrieval Traceability](https://github.com/onestardao/WFGY/blob/main/ProblemMap/retrieval-traceability.md) |
|
||||
| Wrong snippet despite high similarity | Index mismatch after quantization | [Embedding ≠ Semantic](https://github.com/onestardao/WFGY/blob/main/ProblemMap/embedding-vs-semantic.md), [Vectorstore Fragmentation](https://github.com/onestardao/WFGY/blob/main/ProblemMap/vectorstore-fragmentation.md) |
|
||||
| JSON breaks frequently | Quantization noise amplifies schema drift | [Logic Collapse](https://github.com/onestardao/WFGY/blob/main/ProblemMap/logic-collapse.md), [Data Contracts](https://github.com/onestardao/WFGY/blob/main/ProblemMap/data-contracts.md) |
|
||||
| Long chain divergence after 20–40 steps | Numerical error accumulation | [Entropy Collapse](https://github.com/onestardao/WFGY/blob/main/ProblemMap/entropy-collapse.md), [Context Drift](https://github.com/onestardao/WFGY/blob/main/ProblemMap/context-drift.md) |
|
||||
| Deployment mismatch | Torch vs ExLlama kernels version skew | [Bootstrap Ordering](https://github.com/onestardao/WFGY/blob/main/ProblemMap/bootstrap-ordering.md), [Pre-deploy Collapse](https://github.com/onestardao/WFGY/blob/main/ProblemMap/predeploy-collapse.md) |
|
||||
|
||||
---
|
||||
|
||||
## Fix in 60 seconds
|
||||
|
||||
1) **Measure ΔS**
|
||||
Run 20 QA pairs on FP16 baseline vs ExLlamaV2.
|
||||
Acceptable drift ≤ 0.10.
|
||||
|
||||
2) **Probe λ_observe**
|
||||
Increase retrieval k. If λ flips divergent, apply BBAM schema lock.
|
||||
|
||||
3) **Apply the module**
|
||||
- Retrieval drift → BBMC + Retrieval Traceability
|
||||
- Reasoning collapse → BBCR + BBAM clamp
|
||||
- Long-chain instability → BBPF alternate paths
|
||||
|
||||
4) **Verify**
|
||||
Coverage ≥ 0.70, λ convergent, ΔS ≤ 0.10.
|
||||
|
||||
---
|
||||
|
||||
## Minimal setup
|
||||
|
||||
```python
|
||||
from transformers import AutoTokenizer
|
||||
from exllamav2 import ExLlamaV2, ExLlamaV2Cache, ExLlamaV2Tokenizer
|
||||
|
||||
model_path = "your-llama-model"
|
||||
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
||||
|
||||
# Initialize ExLlamaV2
|
||||
model = ExLlamaV2(model_path, quant="4bit", gpu_split="auto")
|
||||
cache = ExLlamaV2Cache(model)
|
||||
|
||||
prompt = "Hello, world!"
|
||||
tokens = tokenizer.encode(prompt, return_tensors="pt").cuda()
|
||||
|
||||
output = model.generate(tokens, max_new_tokens=128, cache=cache)
|
||||
print(tokenizer.decode(output[0]))
|
||||
````
|
||||
|
||||
---
|
||||
|
||||
## Ops checklist
|
||||
|
||||
* Always compare ΔS/λ vs FP16 baseline before shipping
|
||||
* Pin ExLlama kernels to version matching torch/cuBLAS build
|
||||
* Log coverage and citation schema at runtime
|
||||
* Guard JSON outputs with schema validators
|
||||
|
||||
---
|
||||
|
||||
### 🔗 Quick-Start Downloads (60 sec)
|
||||
|
||||
| Tool | Link | 3-Step Setup |
|
||||
| -------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------- |
|
||||
| **WFGY 1.0 PDF** | [Engine Paper](https://github.com/onestardao/WFGY/blob/main/I_am_not_lizardman/WFGY_All_Principles_Return_to_One_v1.0_PSBigBig_Public.pdf) | 1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + \<your question>” |
|
||||
| **TXT OS (plain-text OS)** | [TXTOS.txt](https://github.com/onestardao/WFGY/blob/main/OS/TXTOS.txt) | 1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly |
|
||||
|
||||
---
|
||||
|
||||
### 🧭 Explore More
|
||||
|
||||
| Module | Description | Link |
|
||||
| ------------------------ | ---------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------- |
|
||||
| WFGY Core | WFGY 2.0 engine is live: full symbolic reasoning architecture and math stack | [View →](https://github.com/onestardao/WFGY/tree/main/core/README.md) |
|
||||
| Problem Map 1.0 | Initial 16-mode diagnostic and symbolic fix framework | [View →](https://github.com/onestardao/WFGY/tree/main/ProblemMap/README.md) |
|
||||
| Problem Map 2.0 | RAG-focused failure tree, modular fixes, and pipelines | [View →](https://github.com/onestardao/WFGY/blob/main/ProblemMap/rag-architecture-and-recovery.md) |
|
||||
| Semantic Clinic Index | Expanded failure catalog: prompt injection, memory bugs, logic drift | [View →](https://github.com/onestardao/WFGY/blob/main/ProblemMap/SemanticClinicIndex.md) |
|
||||
| Semantic Blueprint | Layer-based symbolic reasoning & semantic modulations | [View →](https://github.com/onestardao/WFGY/tree/main/SemanticBlueprint/README.md) |
|
||||
| Benchmark vs GPT-5 | Stress test GPT-5 with full WFGY reasoning suite | [View →](https://github.com/onestardao/WFGY/tree/main/benchmarks/benchmark-vs-gpt5/README.md) |
|
||||
| 🧙♂️ Starter Village 🏡 | New here? Lost in symbols? Click here and let the wizard guide you through | [Start →](https://github.com/onestardao/WFGY/blob/main/StarterVillage/README.md) |
|
||||
|
||||
---
|
||||
|
||||
> 👑 **Early Stargazers: [See the Hall of Fame](https://github.com/onestardao/WFGY/tree/main/stargazers)** <img src="https://img.shields.io/github/stars/onestardao/WFGY?style=social" alt="GitHub stars"> ⭐ [WFGY Engine 2.0](https://github.com/onestardao/WFGY/blob/main/core/README.md) is already unlocked. ⭐ Star the repo to help others discover it and unlock more on the [Unlock Board](https://github.com/onestardao/WFGY/blob/main/STAR_UNLOCKS.md).
|
||||
|
||||
<div align="center">
|
||||
|
||||
[](https://github.com/onestardao/WFGY)
|
||||
|
||||
[](https://github.com/onestardao/WFGY/tree/main/OS)
|
||||
|
||||
[](https://github.com/onestardao/WFGY/tree/main/OS/BlahBlahBlah)
|
||||
|
||||
[](https://github.com/onestardao/WFGY/tree/main/OS/BlotBlotBlot)
|
||||
|
||||
[](https://github.com/onestardao/WFGY/tree/main/OS/BlocBlocBloc)
|
||||
|
||||
[](https://github.com/onestardao/WFGY/tree/main/OS/BlurBlurBlur)
|
||||
|
||||
[](https://github.com/onestardao/WFGY/tree/main/OS/BlowBlowBlow)
|
||||
|
||||
</div>
|
||||
Loading…
Add table
Add a link
Reference in a new issue