code wiki / (root) / nx_qwen_blk0_types.nx

nx_qwen_blk0_types.nx

buildroot/runtime/nx_qwen_blk0_types.nx

3604 B85 linesdepth 8pulls 14 transitivereach 0 importersview sourcekind tooltopic qwen
docsdependenciesstructsconstsfunctions

about

nx_qwen_blk0_types.nx -- dump the ggml_type + dims of every blk.0 weight of the real Q4_K_M Qwen. Grounds the integer-layer plan: Q4_K_M mixes quant types per tensor, and our integer linear (nx_q4k_linear) only handles Q4_K (12). This says which blk.0 weights are integer-able (Q4_K=12) vs need the f32 path (Q6_K=14 / F32=0). Metadata parse only (bounded 96MB prefix; tensor infos are at the file start), so no 1.1GB data read. -> /tmp/qwen_blk0_types.txt license_tier: ORIGINAL

dependencies 10 imports · 0 importers

nx_syscalls.nx nx_tier.nx nx_le.nx nx_strconv.nx nx_tensor.nx nx_gguf.nx nx_gguf_load.nx nx_gguf_meta.nx nx_placement.nx nx_gguf_load_lazy.nx nx_qwen_blk0_types.nx

imports: nx_syscalls.nxnx_tier.nxnx_le.nxnx_strconv.nxnx_tensor.nxnx_gguf.nxnx_gguf_load.nxnx_gguf_meta.nxnx_placement.nxnx_gguf_load_lazy.nx

imported by: nobody (leaf or entry point)

call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown

main sys_openat_rd sys_mmap sys_read sys_close nx_gguf_parse nx_le_read_u32 nx_le_read_u64 nx_le_read_u32 ↻ _gguf_skip_value nx_le_read_u64 ↻ nx_le_read_u32 ↻ _gguf_skip_value ↻ sys_mmap ↻ sys_openat_wr ty_dump nx_gguf_find_tensor nx_gguf_tensor_at nx_gguf_name_equals ty_emit sys_mmap ↻ nx_strconv_format_i64 sys_write nx_gguf_tensor_at ↻

structs

none

consts

none

functions

19func ty_emit(fd: i64, name: *u8, name_len: i64, ty: i64, d0: i64, d1: i64) -> i64
called by 1: ty_dump calls 3: sys_mmapnx_strconv_format_i64sys_write
47func ty_dump(fd: i64, hdr: *NxGgufHeader, name: *u8, name_len: i64) -> i64
called by 1: main calls 3: nx_gguf_find_tensorty_emitnx_gguf_tensor_at
54func main() -> i64
calls 7: sys_openat_rdsys_mmapsys_readsys_closenx_gguf_parsesys_openat_wr+1