1pub mod arch_registry;
20pub mod autoregressive;
21pub mod config;
22pub mod dataprocessing;
23pub mod device_capabilities;
24pub mod flow_bridge;
25pub mod flow_util;
26pub mod gguf_config;
27pub mod gguf_resolve;
28pub mod gguf_support;
29pub mod lm;
30pub mod moe_weights;
31pub mod vision_ops_ir;
32pub mod weight_loader;
33pub mod weight_map;
34pub mod weight_registry;
35pub mod weights;
36
37pub use device_capabilities::{
38 STANDARD_DEVICE_NAMES, STANDARD_DEVICES, device_memory_for_moe_offload, is_standard_device,
39 validate_sam_device, validate_standard_device,
40};
41
42pub use gguf_config::{
43 DINOV2_GGUF_ARCHES, EMBED_GGUF_ARCHES, EmbedGgufKind, FLUX_GGUF_ARCHES, GgufMemoryFootprint,
44 SAM_GGUF_ARCHES, SAM2_GGUF_ARCHES, SAM3_GGUF_ARCHES, VJEPA2_GGUF_ARCHES, W2V_BERT_GGUF_ARCHES,
45 embed_gguf_kind, gguf_memory_footprint, gguf_meta_u32, gguf_runner_hint, is_dinov2_gguf_arch,
46 is_embed_gguf_arch, is_flux_gguf_arch, is_sam_gguf_arch, is_sam2_gguf_arch, is_sam3_gguf_arch,
47 is_vjepa2_gguf_arch, is_w2v_bert_gguf_arch,
48};
49pub use gguf_resolve::{
50 GgufTensorNameResolver, LlamaFamilyGgufResolver, PassThroughGgufResolver,
51 PrefixStripGgufResolver, Qwen35NativeGgufResolver, register_gguf_tensor_resolver,
52 resolve_gguf_tensor_name,
53};
54pub use gguf_support::{
55 GgufModelFamily, ResolveWeightsOptions, assert_gguf_family, gguf_architecture_from_path,
56 gguf_architecture_str, gguf_f32_bytes_estimate, gguf_family_for_arch,
57 gguf_safetensors_only_hint, gguf_split_hint, gguf_split_siblings, gguf_validate_arch,
58 list_gguf_files_in_dir, load_gguf_file, resolve_weights_file,
59 resolve_weights_file_with_options,
60};
61
62pub use autoregressive::{
63 KvCacheState, compile_cache_ensure_graph, kv_from_prefill_outputs, past_kv_input_names,
64 prefill_cache_key, run_bucketed_kv_decode, run_bucketed_kv_decode_hir,
65 split_bucketed_decode_kv, split_decode_logits_kv,
66};
67pub use config::{BertConfig, NomicBertConfig, NomicVisionConfig};
68pub use flow_bridge::{
69 apply_compile_profile, compile_graph_encoder, compile_graph_gemma_decode,
70 compile_graph_gemma_prefill, compile_graph_legacy, compile_graph_llama32_decode,
71 compile_graph_llama32_prefill, compile_graph_qwen3_decode, compile_graph_qwen3_prefill,
72 compile_graph_qwen35_decode, compile_graph_qwen35_prefill, compile_graph_sam,
73 compile_graph_with_profile, load_compile_profile, profile_near_weights,
74};
75pub use flow_util::{
76 WeightMapSource, bucket_cache_ensure_built, build_graph, built_from_graph, built_from_hir,
77 built_from_hir_with_profile, compile_built, compile_built_cpu, compile_cache_ensure_built,
78 compile_graph_encoder_with_params, compile_graph_gemma_decode_with_params,
79 compile_graph_gemma_prefill_with_params, compile_graph_profile,
80 compile_graph_qwen3_prefill_with_params, compile_graph_qwen35_decode_with_params,
81 compile_graph_qwen35_prefill_with_params, compile_graph_sam_with_params,
82 compile_graph_with_kv_export_params, graph_from_built, graph_from_hir,
83};
84pub use gguf_resolve::ensure_builtin_resolvers;
85pub use gguf_support::DEFAULT_GGUF_PREFER_SUBSTR;
86pub use lm::{FlowBuildExt, into_compile_parts};
87pub use weight_loader::{
88 GgufLoader, HfTranslatingLoader, WeightLoader, ggml_type_to_quant_scheme, gguf_to_hf_name,
89 gguf_to_hf_name_for_arch, hf_to_gguf_name, is_mtp_weight, load_from_path,
90};
91pub use weight_map::{WeightDrainPolicy, WeightMap};
92pub use weight_registry::{
93 LoadWeightsOptions, LoadedWeights, RegisteredFormat, WeightFormatRegistration,
94 format_for_extension, list_registered_formats, load_weight_map_resolved, load_weights_resolved,
95 open_weight_loader, register_weight_format, registered_extensions_hint,
96};
97pub use weights::{
98 GgufDirGuide, LoadOpts, ResolveOpts, default_resolve_opts, gguf_dir_guide, init,
99 load_weight_map, open, open_map, open_map_with, open_with, pick, pick_default,
100};