메뉴
Maintains half of sliding window as attention sinks with dynamic temporal RoPE alignment plus importance-aware KV cache pruning, enabling 12× extrapolation beyond training length (60+ seconds from 5-second training) without fine-tuning.
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.
SOC 직업 분류 기준
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 deep-forcing-long-video명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SKILL.md 표시 중
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.
| name | deep-forcing-long-video |
| title | Deep Forcing: Training-Free Long Video Generation via Deep Sink and Participative Compression |
| version | 0.0.2 |
| engine | skillxiv-v0.0.2-claude-opus-4.6 |
| license | MIT |
| url | https://arxiv.org/abs/2512.05081 |
| keywords | ["video-generation","long-context","kv-cache-optimization","attention-sinks","temporal-modeling"] |
| description | Maintains half of sliding window as attention sinks with dynamic temporal RoPE alignment plus importance-aware KV cache pruning, enabling 12× extrapolation beyond training length (60+ seconds from 5-second training) without fine-tuning. |
Deep Forcing introduces two mechanisms enabling long video generation without fine-tuning. Deep Sink maintains approximately half of the sliding window as attention sinks while dynamically adjusting temporal RoPE to align sink tokens with current timeline. Participative Compression performs importance-aware KV cache pruning by computing attention scores between recent and candidate tokens. Together these enable 12× extrapolation with maintained visual quality.
Deep Sink Mechanism: Attention sinks are tokens that absorb excess attention mass, preventing context collapse. The key insight is maintaining them at the current timeline:
Sliding window: [sink_1, sink_2, ..., recent_1, recent_2, recent_3]
^--- Half are sinks (fixed at current time)
Dynamic Temporal RoPE: Adjust rotary positional embeddings to align sink positions with current timeline even as the window slides:
rope_angle(sink_t) = current_time - training_length/2
Participative Compression: Prune KV cache by importance:
Sliding window with sinks:
def sliding_window_with_sinks(kv_cache, window_size=256):
sink_ratio = 0.5
sink_size = int(window_size * sink_ratio)
# Maintain sinks at front
sinks = kv_cache[:sink_size]
recent = kv_cache[-sink_size:]
# Concatenate sinks + recent tokens
windowed_kv = torch.cat([sinks, recent], dim=0)
return windowed_kv
Dynamic temporal RoPE:
def dynamic_temporal_rope(positions, current_time, training_length):
# Sink positions: earlier in sequence
sink_positions = torch.arange(len(positions) // 2)
# Adjust sink angles to current time
sink_angles = current_time - (training_length / ) + sink_positions
recent_angles = current_time + torch.arange((positions) // )
combined_angles = torch.cat([sink_angles, recent_angles])
freqs = / ( ** (torch.arange(, dim, ).() / dim))
angles = combined_angles.unsqueeze(-) @ freqs.unsqueeze()
rope = torch.cat([torch.cos(angles), torch.sin(angles)], dim=-)
rope
Importance-aware pruning:
def participative_compression(query, kv_cache, sparsity=0.8):
# Compute attention scores for all KV tokens
scores = query @ kv_cache['keys'].T # [seq_len, kv_len]
# Keep top scoring tokens
keep_ratio = 1.0 - sparsity
num_keep = int(kv_cache['keys'].shape[0] * keep_ratio)
top_k_indices = torch.topk(scores.max(dim=0).values, num_keep).indices
# Prune KV cache
pruned_k = kv_cache['keys'][top_k_indices]
pruned_v = kv_cache['values'][top_k_indices]
return {'keys': pruned_k, 'values': pruned_v}
Long video generation loop:
def generate_long_video(prompt, num_frames=1440): # 60 sec at 24fps
# Initialize from prompt
video = initialize_from_prompt(prompt)
kv_cache = {}
for frame_idx in range(num_frames):
# Apply sliding window + sinks
windowed_kv = sliding_window_with_sinks(kv_cache, window_size=256)
# Apply dynamic temporal RoPE
rope = dynamic_temporal_rope(
positions=torch.arange(windowed_kv.shape[0]),
current_time=frame_idx,
training_length=120 # 5 seconds at 24fps
)
# Apply importance-aware pruning
pruned_kv = participative_compression(
query=video[-1],
kv_cache=windowed_kv,
sparsity=0.8
)
# Generate next frame
next_frame = model(video[-1], pruned_kv, rope)
video = torch.cat([video, next_frame.unsqueeze(0)], dim=0)
# Update cache
kv_cache = update_cache(pruned_kv, next_frame)
return video