Functions — github.com/Mega4alik/ollm

CodeHub MCPcopy Create free account

hub / github.com/Mega4alik/ollm / functions

Functions149 in github.com/Mega4alik/ollm

⨍ Functions 149◇ Types & classes 47

↓ 14 callersFunction_walk_to_parent

Return (parent_obj, leaf_name) for attr_path like 'self_attn.q_proj.weight

src/ollm/utils.py:33

↓ 10 callersMethodset

(self, name, t1)

src/ollm/utils.py:20

↓ 8 callersFunctionfile_get_contents

src/ollm/utils.py:9

↓ 8 callersMethodgenerate

src/ollm/llama.py:185

↓ 8 callersMethodpreload_layer_safetensors

src/ollm/gds_loader.py:354

↓ 7 callersFunction_assign_tensor_to_module

Assign a tensor into target_parent.<leaf>. - If target_parent.<leaf> has a .load call, call it with tensor. - Else, if attribute endswith

src/ollm/utils.py:41

↓ 7 callersFunction_set_meta_placeholder

Replace parameter/module attribute with a tiny meta-device Parameter to free VRAM.

src/ollm/utils.py:72

↓ 7 callersMethodkeys

src/ollm/gds_loader.py:196

↓ 6 callersMethodget_tensor

src/ollm/gds_loader.py:199

↓ 6 callersFunctionrun_test

(test_id, model_id, sm, um, kvcache=None, offload_layers_to_gpu=0, offload_layers_to_cpu=0, max_new_tokens=500

scripts/full_tests.py:5

↓ 5 callersMethod__init__

(self, config: Qwen3NextConfig)

src/ollm/qwen3_next.py:232

↓ 5 callersMethodload_dict_to_cuda

src/ollm/gds_loader.py:284

↓ 5 callersMethodoffload_dict_to_gpu_cpu

(self, base, gpu=False)

src/ollm/gds_loader.py:289

↓ 4 callersMethod__init__

(self, config:Gemma3Config)

src/ollm/gemma3.py:82

↓ 4 callersFunctionget_attn_implementation

src/ollm/inference.py:8

↓ 4 callersMethodini_model

(self, models_dir="./models/", force_download=False)

src/ollm/inference.py:69

↓ 4 callersMethodoffload_layers_to_cpu

src/ollm/inference.py:121

↓ 4 callersMethodprint_and_clean

src/ollm/utils.py:24

↓ 3 callersMethodDiskCache

(self, cache_dir="./kvcache")

src/ollm/inference.py:127

↓ 3 callersMethod_get_my_manifests

src/ollm/gpt_oss.py:85

↓ 3 callersMethod_unload_expert_weights

src/ollm/qwen3_next.py:87

↓ 3 callersMethodget_base

src/ollm/gemma3.py:18

↓ 3 callersFunctionget_optimal_safetensor_reader

(filepath, device=None)

src/ollm/gds_loader.py:258

↓ 3 callersMethodget_seq_length

(self, layer_idx: Optional[int] = 0)

src/ollm/qwen3_next.py:25

↓ 3 callersMethodload_from_disk

(self, layer_idx)

src/ollm/kvcache.py:16

↓ 3 callersMethodload_param_to_cuda

(self, name: str)

src/ollm/gds_loader.py:27

↓ 3 callersMethodsave_to_disk

(self, tensors, layer_idx)

src/ollm/kvcache.py:24

↓ 2 callersMethod__init__

(self, config: GptOssConfig)

src/ollm/gpt_oss.py:124

↓ 2 callersMethod__init__

src/ollm/llama.py:104

↓ 2 callersMethod_layer_param_manifest_names

src/ollm/llama.py:33

↓ 2 callersMethod_load_expert_weights

src/ollm/qwen3_next.py:78

↓ 2 callersMethod_unload_layer_weights

src/ollm/gpt_oss.py:105

↓ 2 callersMethod_unload_layer_weights

src/ollm/qwen3_next.py:72

↓ 2 callersMethod_unload_layer_weights

src/ollm/llama.py:83

↓ 2 callersMethod_unload_layer_weights

src/ollm/gemma3.py:33

↓ 2 callersMethodclose

src/ollm/gds_loader.py:193

↓ 2 callersFunctionconvert_moe_packed_tensors

Convert the mxfp4 weights again, dequantizing and makes them compatible with the forward pass of GPT_OSS.

src/ollm/gds_loader.py:128

↓ 2 callersMethodforward

( self, input_ids: Optional[torch.LongTensor] = None, attention_mask: Optional[torch.Tensor] = None, p

src/ollm/gpt_oss.py:130

↓ 2 callersMethodforward

src/ollm/gemma3.py:42

↓ 2 callersMethodini_ocache

(self, cache_dir, device, stats)

src/ollm/kvcache.py:7

↓ 2 callersMethodload_dict_from_disk

(self, base, device='cpu')

src/ollm/gds_loader.py:301

↓ 2 callersMethodload_from_disk_to_cuda

(self, path, shape, dtype)

src/ollm/gds_loader.py:50

↓ 2 callersMethodload_model

(self, model_dir)

src/ollm/inference.py:84

↓ 2 callersMethodoffload_param_to_cpu

src/ollm/gds_loader.py:85

↓ 1 callersMethod__init__

(self, config, layer_idx: int)

src/ollm/voxtral.py:37

↓ 1 callersMethod_load_experts_weights

(self, experts_idx)

src/ollm/qwen3_next.py:93

↓ 1 callersMethod_load_layer_weights

src/ollm/voxtral.py:19

↓ 1 callersMethod_load_layer_weights

src/ollm/gpt_oss.py:94

↓ 1 callersMethod_load_layer_weights

src/ollm/qwen3_next.py:62

↓ 1 callersMethod_load_layer_weights

src/ollm/llama.py:72

↓ 1 callersMethod_load_layer_weights

src/ollm/gemma3.py:22

↓ 1 callersMethod_unload_experts_weights

(self, experts_idx)

src/ollm/qwen3_next.py:97

↓ 1 callersMethod_unload_layer_weights

src/ollm/voxtral.py:29

↓ 1 callersFunctionattention_ref

( query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, sinks: torch.Tensor, sm_sca

src/ollm/gpt_oss_attention.py:170

↓ 1 callersMethodclose

src/ollm/gds_loader.py:228

↓ 1 callersFunctionconvert_moe_packed_tensors

Convert the mxfp4 weights again, dequantizing and makes them compatible with the forward pass of GPT_OSS.

scripts/gpt_oss_export.py:87

↓ 1 callersMethoddownload_and_unpack

(self, models_dir: str)

src/ollm/inference.py:24

↓ 1 callersMethodforward

src/ollm/qwen3_next.py:103

↓ 1 callersMethodforward

src/ollm/llama.py:20

↓ 1 callersMethodget_dtype

src/ollm/gds_loader.py:40

↓ 1 callersMethodget_offloaded_dict_to_cuda

src/ollm/gds_loader.py:293

↓ 1 callersMethodget_offloaded_from_cpu_to_cuda

src/ollm/gds_loader.py:94

↓ 1 callersFunctionget_sample

scripts/test_autoinference.py:7

↓ 1 callersMethodhf_download

(self, model_dir)

src/ollm/inference.py:53

↓ 1 callersFunctionhf_push_to_hub

scripts/scripts.py:35

↓ 1 callersFunctioninference_chat

scripts/test.py:26

↓ 1 callersFunctionini_model

scripts/test.py:10

↓ 1 callersMethodis_sharded

(self, model_dir)

src/ollm/inference.py:164

↓ 1 callersMethodload_mxfp4_from_disk

(self, path, shape, dtype)

src/ollm/gds_loader.py:80

↓ 1 callersMethodload_torch_from_disk

src/ollm/gds_loader.py:76

↓ 1 callersMethodoffload_layers_to_gpu_cpu

(self, gpu_layers_num=0, cpu_layers_num=0)

src/ollm/qwen3_next.py:330

↓ 1 callersMethodoffload_layers_to_gpu_cpu

src/ollm/inference.py:124

↓ 1 callersFunctiononline_chunked_grouped_attention_rope

Online chunked grouped multi-head attention (RoPE already applied to q,k). - q: queries (B, Hq, Lq, D) - k,v: keys/values with Hkv heads

src/ollm/attention.py:141

↓ 1 callersMethodupdate

( self, key_states: torch.Tensor, value_states: torch.Tensor, layer_idx: int, cache_kwargs: Optional

src/ollm/kvcache.py:37

Method__enter__

src/ollm/gds_loader.py:222

(self, exc_type, exc_val, exc_tb)

src/ollm/gds_loader.py:225

Method__getitem__

(self, layer_idx: int)

src/ollm/qwen3_next.py:28

src/ollm/voxtral.py:69

(self, config: GptOssConfig, layer_idx: int)

src/ollm/gpt_oss.py:112

src/ollm/gpt_oss.py:267

(self, config, cache_dir="./kv_cache", device="cuda:0", stats=None)

src/ollm/qwen3_next.py:20

(self, config, layer_idx)

src/ollm/qwen3_next.py:214

src/ollm/qwen3_next.py:318

src/ollm/qwen3_next.py:349

src/ollm/qwen3_next.py:377

(self, cache_dir="./kv_cache", device="cuda:0", stats=None)

src/ollm/kvcache.py:33

src/ollm/utils.py:17

(self, config: LlamaConfig, layer_idx: int)

src/ollm/llama.py:92

src/ollm/llama.py:199

(self, config, layer_idx)

src/ollm/gemma3.py:55

(self, config: Gemma3Config)

src/ollm/gemma3.py:67

src/ollm/gemma3.py:108

src/ollm/gemma3.py:114

(self, path: str, device="cuda:0")

src/ollm/gds_loader.py:19

src/ollm/gds_loader.py:184

(self, path: str, device="cuda:0")

src/ollm/gds_loader.py:210

(self, path: str, device="cuda:0")

src/ollm/gds_loader.py:267

(self, path: str, device="cuda:0")

src/ollm/gds_loader.py:315

(self, path: str, device="cuda:0")

src/ollm/gds_loader.py:337

(self, model_id, device="cuda:0", logging=True, multimodality=False)

src/ollm/inference.py:18

next →1–100 of 149, ranked by callers