rlx_models_core/
lib.rs

1// RLX — versatile ML compiler + runtime.
2// Copyright (C) 2026 Eugene Hauptmann, Nataliya Kosmyna.
3//
4// This program is free software: you can redistribute it and/or modify
5// it under the terms of the GNU General Public License as published by
6// the Free Software Foundation, version 3.
7//
8// This program is distributed in the hope that it will be useful,
9// but WITHOUT ANY WARRANTY; without even the implied warranty of
10// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11// GNU General Public License for more details.
12//
13// You should have received a copy of the GNU General Public License
14// along with this program. If not, see <https://www.gnu.org/licenses/>.
15
16//! Shared infrastructure for RLX model crates: HuggingFace config parsing,
17//! safetensors / GGUF weight loading, tier-1 compile profile helpers, and
18//! packed GGUF prefill guards ([`flow_bridge::packed_gguf_compile_guard`], etc.).
19
20pub mod arch_registry;
21pub mod asr_bench;
22pub mod asr_metrics;
23pub mod audio;
24pub mod audio_codec;
25pub mod audio_ops_ir;
26pub mod autoregressive;
27pub mod codec_bench;
28pub mod config;
29pub mod dataprocessing;
30pub mod device_capabilities;
31pub mod embedded_safetensors;
32pub mod flow_bridge;
33pub mod flow_util;
34pub mod gguf_config;
35pub mod gguf_resolve;
36pub mod gguf_support;
37pub mod gpu_kv;
38pub mod host_kernels;
39pub mod image_preprocess;
40pub mod lm;
41pub mod moe_weights;
42pub mod prompt_cache;
43pub mod safetensors_checkpoint;
44pub mod vision_ops_ir;
45pub mod weight_loader;
46pub mod weight_map;
47pub mod weight_registry;
48pub mod weights;
49
50pub use asr_metrics::{
51    EditCounts, WerAccumulator, batch_to_stream_factor, character_error_rate, edit_distance,
52    normalize_words, rtfx, word_edit_counts, word_error_rate,
53};
54pub use audio::{resample_linear, resample_linear_interleaved};
55pub use audio_codec::{
56    AudioCodec, ChunkStreamer, CodecInfo, CompressStats, FileCodec, HierarchicalCodes, RvqCodes,
57};
58pub use device_capabilities::{
59    LM_DEVICE_NAMES, LM_INFERENCE_DEVICE_PRIORITY, STANDARD_DEVICE_NAMES, STANDARD_DEVICES,
60    device_memory_for_moe_offload, is_lm_device, is_standard_device, pick_lm_device,
61    resolve_lm_device_str, validate_lm_device, validate_sam_device, validate_standard_device,
62};
63
64pub use gguf_config::{
65    DINOV2_GGUF_ARCHES, EMBED_GGUF_ARCHES, EmbedGgufKind, FLUX_GGUF_ARCHES, GgufMemoryFootprint,
66    SAM_GGUF_ARCHES, SAM2_GGUF_ARCHES, SAM3_GGUF_ARCHES, VJEPA2_GGUF_ARCHES, W2V_BERT_GGUF_ARCHES,
67    embed_gguf_kind, gguf_memory_footprint, gguf_meta_u32, gguf_runner_hint, is_dinov2_gguf_arch,
68    is_embed_gguf_arch, is_flux_gguf_arch, is_sam_gguf_arch, is_sam2_gguf_arch, is_sam3_gguf_arch,
69    is_vjepa2_gguf_arch, is_w2v_bert_gguf_arch,
70};
71pub use gguf_resolve::{
72    GgufTensorNameResolver, LlamaFamilyGgufResolver, PassThroughGgufResolver,
73    PrefixStripGgufResolver, Qwen35NativeGgufResolver, register_gguf_tensor_resolver,
74    resolve_gguf_tensor_name,
75};
76pub use gguf_support::{
77    GgufModelFamily, ResolveWeightsOptions, assert_gguf_family, gguf_architecture_from_path,
78    gguf_architecture_str, gguf_f32_bytes_estimate, gguf_family_for_arch,
79    gguf_safetensors_only_hint, gguf_split_hint, gguf_split_siblings, gguf_validate_arch,
80    list_gguf_files_in_dir, load_gguf_file, resolve_weights_file,
81    resolve_weights_file_with_options,
82};
83
84pub use autoregressive::{
85    KvCacheState, compact_bucketed_kv_buffer, compile_cache_ensure_graph, infer_prefill_kv_seq,
86    kv_from_prefill_outputs, kv_from_prefill_outputs_per_layer,
87    packed_prefill_active_extent_enabled, past_kv_input_names, prefill_cache_key,
88    run_bucketed_kv_decode, run_bucketed_kv_decode_graph_layers_scratch,
89    run_bucketed_kv_decode_hir, run_bucketed_kv_decode_hir_scratch,
90    run_bucketed_kv_decode_hir_uniform, run_bucketed_kv_decode_keyed,
91    run_bucketed_kv_decode_keyed_batched, run_packed_prefill, split_bucketed_decode_kv,
92    split_bucketed_decode_kv_per_layer, split_decode_logits_kv, split_decode_logits_kv_aux,
93};
94pub use config::{BertConfig, NomicBertConfig, NomicVisionConfig};
95pub use embedded_safetensors::EmbeddedSafetensors;
96pub use flow_bridge::{
97    apply_compile_profile, compile_graph_encoder, compile_graph_gemma_decode,
98    compile_graph_gemma_prefill, compile_graph_legacy, compile_graph_llama32_decode,
99    compile_graph_llama32_prefill, compile_graph_qwen3_decode, compile_graph_qwen3_prefill,
100    compile_graph_qwen35_decode, compile_graph_qwen35_prefill, compile_graph_sam,
101    compile_graph_with_profile, compile_options_for_packed_gguf_prefill,
102    compile_options_for_packed_gguf_prefill_with_profile, compile_options_for_profile,
103    load_compile_profile, packed_gguf_compile_guard, packed_gguf_execution_device,
104    profile_near_weights,
105};
106pub use flow_util::{
107    WeightMapSource, bucket_cache_ensure_built, build_graph, built_from_graph, built_from_hir,
108    built_from_hir_with_profile, compile_built, compile_built_cpu, compile_cache_ensure_built,
109    compile_cache_ensure_built_with_options, compile_graph_encoder_with_params,
110    compile_graph_gemma_decode_with_params, compile_graph_gemma_prefill_with_params,
111    compile_graph_profile, compile_graph_qwen3_prefill_with_params,
112    compile_graph_qwen35_decode_with_params, compile_graph_qwen35_prefill_with_params,
113    compile_graph_sam_with_params, compile_graph_with_kv_export_params, graph_from_built,
114    graph_from_hir,
115};
116pub use gguf_resolve::ensure_builtin_resolvers;
117pub use gguf_support::DEFAULT_GGUF_PREFER_SUBSTR;
118pub use gpu_kv::{
119    GpuKvBinding, GpuKvCacheSet, cross_attn_gpu_handles_ready, device_supports_gpu_kv,
120    install_cross_attn_gpu_handles, install_gpu_kv_handles, reinstall_gpu_kv_handles,
121    run_bucketed_kv_decode_gpu, run_bucketed_kv_decode_gpu_hir, run_bucketed_kv_mtp_gpu,
122    sync_gpu_kv_to_host,
123};
124pub use lm::{FlowBuildExt, into_compile_parts};
125pub use safetensors_checkpoint::SafetensorsCheckpoint;
126pub use weight_loader::{
127    ArcCacheLoader, ArcF32Tensor, GgufLoader, HfTranslatingLoader, WeightLoader,
128    dequant_matmul_supported, ggml_type_to_quant_scheme, gguf_to_hf_name, gguf_to_hf_name_for_arch,
129    hf_to_gguf_name, hf_to_gguf_name_for_arch, is_mtp_weight, load_from_path,
130};
131pub use weight_map::{WeightDrainPolicy, WeightMap};
132pub use weight_registry::{
133    LoadWeightsOptions, LoadedWeights, RegisteredFormat, WeightFormatRegistration,
134    format_for_extension, list_registered_formats, load_weight_map_resolved, load_weights_resolved,
135    open_weight_loader, register_weight_format, registered_extensions_hint,
136};
137pub use weights::{
138    GgufDirGuide, LoadOpts, ResolveOpts, default_resolve_opts, gguf_dir_guide, init,
139    load_weight_map, open, open_map, open_map_with, open_with, pick, pick_default,
140};
rlx_models_core/lib.rs

rlx_models_core/
lib.rs
