llama.cpp src/llama-quant.cpp (llama_tensor_get_type_impl)

Fields

id

sources/qlab2-llama-quant-cpp

URL

https://github.com/ggml-org/llama.cpp/blob/master/src/llama-quant.cpp

Publisher

ggml-org

Kind

repo

Accessed

2026-10-03

notes

Authoritative source for the _S/_M/_L MIXTURE rules: which tensor in which layer gets which ggml type for each ftype. Contains use_more_bits() = i_layer < n/8 || i_layer >= 7n/8 || (i_layer - n/8)%3==2, the ATTENTION_V / FFN_DOWN / ATTENTION_OUTPUT / ATTENTION_QKV / FFN_GATE / FFN_UP special cases, the OUTPUT-tensor bump rules, the n_expert==8 (Mixtral) rules, and llama_tensor_get_quant_type() which shows LLAMA_FTYPE_MOSTLY_IQ2_S -> GGML_TYPE_IQ2_XS, IQ2_M -> IQ2_S, IQ3_XS -> IQ3_S, IQ3_M -> IQ3_S. | SOURCE AUDIT 2026-10-03: the 'published' field held a verbatim copy of this notes paragraph instead of a date (field corruption across the qlab2-* batch); published set to null because a GitHub README, doc page or source header has no reliable publication date of its own. Content of the document itself was re-verified live 2026-10-03 and the notes are correct.

Notes

Authoritative source for the _S/_M/_L MIXTURE rules: which tensor in which layer gets which ggml type for each ftype. Contains use_more_bits() = i_layer < n/8 || i_layer >= 7n/8 || (i_layer - n/8)%3==2, the ATTENTION_V / FFN_DOWN / ATTENTION_OUTPUT / ATTENTION_QKV / FFN_GATE / FFN_UP special cases, the OUTPUT-tensor bump rules, the n_expert==8 (Mixtral) rules, and llama_tensor_get_quant_type() which shows LLAMA_FTYPE_MOSTLY_IQ2_S -> GGML_TYPE_IQ2_XS, IQ2_M -> IQ2_S, IQ3_XS -> IQ3_S, IQ3_M -> IQ3_S. | SOURCE AUDIT 2026-10-03: the 'published' field held a verbatim copy of this notes paragraph instead of a date (field corruption across the qlab2-* batch); published set to null because a GitHub README, doc page or source header has no reliable publication date of its own. Content of the document itself was re-verified live 2026-10-03 and the notes are correct.

Cross-references

This record references no other records.