Unverified50% confidenceFactExact time
Hy-Memory-30B-A3B uses the HYV3ForCausalLM architecture, with 128 routed experts and 8 activated per token, 48 layers, hidden dimension 2048, max position embeddings 262,144, and support for Chinese and English.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn vLLM v0.31.0 testing, vllm serve tencent/Hy-Memory-30B-A3B --served-model-name hy-memory can start the BF16 checkpoint, with weights of about 60 GB, requiring one 80 GB card or two 48 GB cards with tensor parallelism.69% similarUnverifiedHy-Memory-30B-A3B itself does not store memories; instead it acts as a memory writer, specifically screening information to write into an external memory store and updating old entries.65% similarUnverifiedTaking the LLaMA-3 70B model as an example, with 80 layers, head_dim=128, and BF16 precision, each token occupies approximately 80KB of KV cache.63% similarUnverifiedA100/H100等数据中心GPU采用HBM高带宽内存技术，通过3D堆叠实现80GB乃至更大显存容量59% similarUnverifiedJEV-27B-VL exposes two distinct interfaces: System 1 reads logits and returns typed decisions and calibrated probabilities, supporting yes/no judgments, selection among 2 to 256 options, and scoring from 0 to 5, with prompts up to 256K tokens; System 2 is described as unmodified Qwen3.8-27B, which accepts image input and outputs text or step-by-step reasoning.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1054161API
curl https://kongchang.com/api/v1/knowledge/claims/1054161MCP
get_claim(id=1054161)