Unverified50% confidenceFactExact time
In 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.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
UnverifiedMiniMax H3's official documentation requires ≥64 GB RAM; three of the four test machines peaked at 68 GB of memory usage, making memory a hard bottleneck for running on consumer-grade GPUs73% similarUnverifiedVLM-Seek 10B can be run by distributing the model layers across dual 4060Ti 16GB graphics cards, but this requires modifying the original inference code and involves a more complex configuration.71% similarUnverifiedThe VLM-Seek 1.5 10B model weights themselves occupy approximately 20GB of VRAM.71% similarUnverifiedHy-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.69% similarUnverifiedFor local deployment of MiniMax H3, virtual memory must be set to 128 GB or more, which is a hard requirement for ensuring stable model operation.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1054169API
curl https://kongchang.com/api/v1/knowledge/claims/1054169MCP
get_claim(id=1054169)