메뉴
Extract high-quality embeddings from frozen decoder-only LLMs by re-routing internal key-value states without training—outperforming training-free baselines by 10% on MTEB while maintaining robustness across sequences up to 4,096 tokens.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/ADu2021/skillXiv --skill kv-embedding-training-free명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
Convert arXiv papers into ready-to-use agent skills using category-aware extraction. First classifies the paper into one or more of 11 research categories, then applies a specialized extraction pipeline for each category — because different types of papers produce different types of usable knowledge. A single paper can yield multiple skills if it spans categories. Use this skill whenever the user wants to turn a paper into a skill, extract practical techniques from research, build a skill library from papers, convert arXiv papers into reusable agent instructions, or batch-process multiple papers into skills. Also trigger when someone asks about extracting actionable knowledge from papers, making research practical for LLM agents, or systematically converting academic contributions into structured agent capabilities.
Establish regret bounds for behavior cloning with discretized actions combining statistical error and quantization error terms. Prove smoothness requirements for safe quantizer design, show that learning-based quantizers fail these requirements, and propose model-based augmentation to reduce error dependence from H² to H.
Dynamically allocate LoRA ranks per-layer during fine-tuning instead of using fixed uniform ranks. Learn optimal rank for each layer and subject via variational framework with discretized exponential distribution, reducing memory footprint while maintaining fidelity and text-alignment.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | kv-embedding-training-free |
| title | KV-Embedding: Training-free Text Embedding via Internal KV Re-routing in Decoder-only LLMs |
| version | 0.0.2 |
| engine | skillxiv-v0.0.2-claude-opus-4.6 |
| license | MIT |
| url | https://arxiv.org/abs/2601.01046 |
| keywords | ["Text Embeddings","Decoder-only Models","Representation Learning","KV Caching","Model Internals"] |
| description | Extract high-quality embeddings from frozen decoder-only LLMs by re-routing internal key-value states without training—outperforming training-free baselines by 10% on MTEB while maintaining robustness across sequences up to 4,096 tokens. |
KV-Embedding solves a critical limitation: while decoder-only LLMs (Qwen, Mistral, Llama) excel at generation, extracting semantic embeddings from them is challenging. Their causal attention masks early tokens from later context, and their objective (next-token prediction) biases representations toward generation over semantic compression.
Core Innovation: Leverage internal key-value (KV) cache states that naturally encode sequence-level information. By re-routing these states as prepended prefixes, enable all tokens to access full sequence context within a single forward pass—activating latent embedding capabilities without training.
Causal Attention Limitation: Standard decoder-only attention masks are causal—early tokens cannot attend to future tokens. This is efficient for generation but problematic for embeddings.
Next-Token Prediction Bias: Models optimized for generation focus on predicting the next token. Embeddings trained this way emphasize generative features (fluency, diversity) over semantic compression.
Training Complexity: Typical embedding approaches require fine-tuning with contrastive losses, adding computational overhead and reducing accessibility.
During forward pass, capture key-value cache states from each transformer layer:
def extract_kv_states(model, input_ids):
"""Extract KV cache states during forward pass."""
hidden_states = []
with torch.no_grad():
for layer_idx, layer in enumerate(model.model.layers):
# Forward through layer, capturing KV cache
out, cache = layer(
input_ids,
attention_mask=...,
past_key_values=...
)
hidden_states.append(cache)
# KV states from final layer contain full sequence information
return hidden_states[-1]
Use final layer's KV states as a prepended prefix in a second forward pass:
Key Insight: The KV states of the final token encode a compressed view of the entire sequence. By prepending these states, all tokens gain access to sequence-level context.
def rerout_kv_as_prefix(model, input_ids):
"""Re-route KV states as prefix for embedding extraction."""
# Forward pass 1: Extract KV states
kv_states = extract_kv_states(model, input_ids)
# Prepare prefix: prepend KV states
# This enables causal attention to access full sequence context
prefix_kv = kv_states # Use final token's KV cache as prefix
# Forward pass 2: Generate with prefix
# Now all positions can access full sequence via prefix
embeddings = model(
input_ids,
past_key_values=prefix_kv,
use_cache=False, # Don't generate, just forward
)
return embeddings
Different layers encode different information levels. Automatically select optimal layers:
Intrinsic Dimensionality: A metric indicating how much useful information is encoded in a representation.
def select_optimal_layers(model, validation_texts):
"""Determine which layers provide best embeddings."""
layer_scores = {}
for layer_idx in range(num_layers):
# Extract representations from specific layer
representations = [
extract_layer_representation(text, layer_idx)
for text in validation_texts
]
# Compute intrinsic dimensionality
dim = compute_intrinsic_dimensionality(representations)
# Score based on dimension and downstream task performance
score = evaluate_on_retrieval_task(representations)
layer_scores[layer_idx] = score
# Select layers with best scores
best_layers = sorted(
layer_scores.items(), key=lambda x: x[1], reverse=True
)[:num_select]
return [layer_idx for layer_idx, _ in best_layers]
Model-Agnostic: Automatically selects layers for Qwen, Mistral, Llama without manual tuning.
MTEB Benchmark Results:
Sequence Length Robustness:
Computational Efficiency:
vs. Fine-tuned Embeddings:
vs. Simple Pooling Methods:
vs. Contrastive Learning:
Model Compatibility: Works with any HuggingFace decoder-only model with accessible KV cache:
compatible_models = [
"Qwen/Qwen2-7B",
"mistralai/Mistral-7B-v0.1",
"meta-llama/Llama-2-7b",
"meta-llama/Llama-3-8B",
# And many more...
]
Dimension Selection: Automated layer selection works across model families. For new model architectures:
def auto_configure_for_model(model_name):
"""Automatically configure KV-Embedding for new model."""
model = AutoModel.from_pretrained(model_name)
# Validate on small subset
validation_texts = load_validation_set()
# Run automated layer selection
optimal_layers = select_optimal_layers(model, validation_texts)
dimension = optimal_layers[0] # Use best layer's dimension
return {"layers": optimal_layers, "dimension": dimension}
Use when:
When NOT to use:
Why KV States Encode Semantics:
Implementation available with examples for major model families.
Supported Models: