IsValorum/MiniCPM-V-4.7-35B-A3B-APEX-I-MiniPlus-V2.1-GGUF

Quick Navigation Index

Independently computed using dedicated cloud compute clusters, funded 100% out-of-pocket.

If you consider these releases to be of high quality and they help you in your day-to-day workflow, consider supporting on Ko-fi. Every contribution directly sustains future cloud compute hours and helps support my university studies.

---

MiniCPM-V 4.7 35B-A3B APEX-I MiniPlus V2.1 GGUF

Official APEX-I MiniPlus V2.1 canonical release for OpenBMB MiniCPM-V 4.7 35B-A3B, the first Mixture-of-Experts (MoE) multimodal model in the MiniCPM series.

Key Architectural Features:

• Canonical 733-Tensor Allocation:
  • output.weight protected in Q6_K (1 tensor).
  • token_embd.weight protected in Q4_K (1 tensor).
  • Normalizations and MoE Expert Routers (ffn_gate_inp) in F32 (zero routing drift).
  • Shared foundation experts (shexp) in Q5_K (100% active pathway).
  • Periodic full-attention anchors (layers 3, 7, 11, ..., 39) in Q4_K (Q/K/V) and Q6_K (output).
  • Recurrent Gated DeltaNet protected states in F32 (ssm_a, ssm_conv1d, ssm_dt, ssm_norm).
  • Edge routed experts (layers 0-9 and 30-39) in Q3_K; core routed experts (layers 10-29) in IQ3_XXS.
• Multimodal Projector:
  • Includes companion mmproj-Q8_0.gguf for high-fidelity visual reasoning without token degradation.

CLI Usage Example:

llama-minicpmv-cli \
  -m MiniCPM-V-4.7-35B-A3B.APEX-I-MiniPlus-V2.1.gguf \
  --mmproj mmproj-Q8_0.gguf \
  --image test.jpg \
  -p "Describe what you see in this image in detail." \
  -c 4096 --temp 0.7