DeepSeek-V4-Flash-Spark-Agentic — superseded

This id is redirect-only.

Use the canonical experimental pack:

• HF: https://huggingface.co/hizrianraz/DeepSeek-V4-Flash-REAP25-Spark-Agentic
• GH: https://github.com/hizrianraz/DeepSeek-V4-Flash-REAP25-Spark-Agentic

Reference facts — not parameters hosted here

Total (upstream)	284B (official DeepSeek card)
Active / token	13B
Context	1M tokens
Arch	MoE · DeepseekV4ForCausalLM
Canonical docs path	REAP25 pack — not this id
This repo	redirect only — no weights · no serve · no measure

The upstream figures are reference facts, not measurements or files from this repository. This surface does not host weights, quants, or a runnable serve path.

Why this exists

Legacy non-REAP pack id kept so old links do not 404.

No weights. No serve path. No measure claim. No runnable pull/serve from this repo.

Family (Aug 3)

1. Laguna-S-2.1-Spark-Agentic — historical measured; freeze-attested; non-hero until host receipt
2. Qwen3-Coder-Next-Spark-Agentic — FP8 Preview; exact-runtime evidence pending
3. DeepSeek-V4-Flash-REAP25-Spark-Agentic — HOLD / NO_HERO experimental scaffold (canonical DeepSeek id)

Related repositories

• Base: deepseek-ai/DeepSeek-V4-Flash
• Canonical pack (use this): hizrianraz/DeepSeek-V4-Flash-REAP25-Spark-Agentic
• This id: redirect / supersession pointer only

Upstream reference (not this pack)

Upstream DeepSeek-V4-Flash card numbers (reference only):

• ~284B total · 13B active/token · 1M context · MoE
• No size, fit, residency, context, speed, or stability conclusion is made here.
• The pinned third-party artifact and its limitations are documented only on the REAP25 pack id.

Honesty

• Redirect ≠ second ship path
• Co-list ≠ measured ship
• Verifier ≠ gate clearance
• Personal experimental work — not an Ainfera product
• Upstream: deepseek-ai/DeepSeek-V4-Flash (MIT) — not affiliated with DeepSeek
• NO_HERO · diy_gguf=false · promo_pre=false

License / notice

See LICENSE and NOTICE for redirect legal surface only.