anodyine/vllm-omni
1
0 Fork 0
Code Issues 4 Pull Requests 2 Actions Packages Projects Releases Wiki Activity

[Tests] Config driven tiny model builder (#5090)

Browse Source 
Signed-off-by: Nick Cao <ncao@redhat.com>
This commit is contained in:
Nick Cao authored and GitHub committed 2026-07-22 17:11:53 +00:00
1 parent cc1939120f
commit f10320f53f
18 files changed
+601 -60
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

tests/model_tests/diffusion/tiny_configs/Flux2KleinPipeline/model_index.json
+25
View File
@@ -0,0 +1,25 @@
{
"_class_name": "Flux2KleinPipeline",
"_diffusers_version": "0.37.0.dev0",
"is_distilled": true,
"scheduler": [
"diffusers",
"FlowMatchEulerDiscreteScheduler"
],
"text_encoder": [
"transformers",
"Qwen3ForCausalLM"
],
"tokenizer": [
"transformers",
"Qwen2TokenizerFast"
],
"transformer": [
"diffusers",
"Flux2Transformer2DModel"
],
"vae": [
"diffusers",
"AutoencoderKLFlux2"
]
}
tests/model_tests/diffusion/tiny_configs/Flux2KleinPipeline/text_encoder/config.json
+60
View File
@@ -0,0 +1,60 @@
{
"architectures": [
"Qwen3ForCausalLM"
],
"attention_bias": false,
"attention_dropout": 0.0,
"bos_token_id": 151643,
"dtype": "bfloat16",
"eos_token_id": 151645,
"head_dim": 16,
"hidden_act": "silu",
"hidden_size": 32,
"initializer_range": 0.02,
"intermediate_size": 64,
"layer_types": [
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention",
"full_attention"
],
"max_position_embeddings": 512,
"max_window_layers": 36,
"model_type": "qwen3",
"num_attention_heads": 2,
"num_hidden_layers": 28,
"num_key_value_heads": 2,
"rms_norm_eps": 1e-06,
"rope_scaling": null,
"rope_theta": 1000000,
"sliding_window": null,
"tie_word_embeddings": true,
"transformers_version": "4.56.1",
"use_cache": true,
"use_sliding_window": false,
"vocab_size": 151936
}
tests/model_tests/diffusion/tiny_configs/Flux2KleinPipeline/transformer/config.json
+24
View File
@@ -0,0 +1,24 @@
{
"_class_name": "Flux2Transformer2DModel",
"_diffusers_version": "0.37.0.dev0",
"_name_or_path": "/raid/yiyi/klein-4b-distilled-diffusers/transformer",
"attention_head_dim": 16,
"axes_dims_rope": [
4,
4,
4,
4
],
"eps": 1e-06,
"guidance_embeds": false,
"in_channels": 64,
"joint_attention_dim": 96,
"mlp_ratio": 3.0,
"num_attention_heads": 2,
"num_layers": 2,
"num_single_layers": 2,
"out_channels": null,
"patch_size": 1,
"rope_theta": 2000,
"timestep_guidance_channels": 32
}
tests/model_tests/diffusion/tiny_configs/Flux2KleinPipeline/vae/config.json
+31
View File
@@ -0,0 +1,31 @@
{
"_class_name": "AutoencoderKLFlux2",
"_diffusers_version": "0.37.0.dev0",
"_name_or_path": "black-forest-labs/FLUX.2-dev",
"act_fn": "silu",
"batch_norm_eps": 0.0001,
"batch_norm_momentum": 0.1,
"block_out_channels": [
32
],
"down_block_types": [
"DownEncoderBlock2D"
],
"force_upcast": true,
"in_channels": 3,
"latent_channels": 16,
"layers_per_block": 1,
"mid_block_add_attention": true,
"norm_num_groups": 16,
"out_channels": 3,
"patch_size": [
2,
2
],
"sample_size": 1024,
"up_block_types": [
"UpDecoderBlock2D"
],
"use_post_quant_conv": true,
"use_quant_conv": true
}
tests/model_tests/diffusion/tiny_configs/LTX2Pipeline/audio_vae/config.json
+25
View File
@@ -0,0 +1,25 @@
{
"_class_name": "AutoencoderKLLTX2Audio",
"_diffusers_version": "0.37.0.dev0",
"attn_resolutions": null,
"base_channels": 128,
"causality_axis": "height",
"ch_mult": [
1,
2,
4
],
"double_z": true,
"dropout": 0.0,
"in_channels": 2,
"is_causal": true,
"latent_channels": 8,
"mel_bins": 64,
"mel_hop_length": 160,
"mid_block_add_attention": false,
"norm_type": "pixel",
"num_res_blocks": 1,
"output_channels": 2,
"resolution": 256,
"sample_rate": 16000
}
tests/model_tests/diffusion/tiny_configs/LTX2Pipeline/connectors/config.json
+21
View File
@@ -0,0 +1,21 @@
{
"_class_name": "LTX2TextConnectors",
"_diffusers_version": "0.37.0.dev0",
"audio_connector_attention_head_dim": 16,
"audio_connector_num_attention_heads": 2,
"audio_connector_num_layers": 1,
"audio_connector_num_learnable_registers": 4,
"caption_channels": 32,
"causal_temporal_positioning": false,
"connector_rope_base_seq_len": 32,
"rope_double_precision": true,
"rope_theta": 10000.0,
"rope_type": "split",
"text_proj_in_factor": 3,
"video_connector_attention_head_dim": 16,
"video_connector_num_attention_heads": 2,
"video_connector_num_layers": 1,
"video_connector_num_learnable_registers": 4,
"video_hidden_dim": 32,
"audio_hidden_dim": 32
}
tests/model_tests/diffusion/tiny_configs/LTX2Pipeline/model_index.json
+36
View File
@@ -0,0 +1,36 @@
{
"_class_name": "LTX2Pipeline",
"_diffusers_version": "0.37.0.dev0",
"audio_vae": [
"diffusers",
"AutoencoderKLLTX2Audio"
],
"connectors": [
"ltx2",
"LTX2TextConnectors"
],
"scheduler": [
"diffusers",
"FlowMatchEulerDiscreteScheduler"
],
"text_encoder": [
"transformers",
"Gemma3ForConditionalGeneration"
],
"tokenizer": [
"transformers",
"GemmaTokenizerFast"
],
"transformer": [
"diffusers",
"LTX2VideoTransformer3DModel"
],
"vae": [
"diffusers",
"AutoencoderKLLTX2Video"
],
"vocoder": [
"ltx2",
"LTX2Vocoder"
]
}
tests/model_tests/diffusion/tiny_configs/LTX2Pipeline/text_encoder/config.json
+68
View File
@@ -0,0 +1,68 @@
{
"architectures": [
"Gemma3ForConditionalGeneration"
],
"boi_token_index": 255999,
"dtype": "float32",
"eoi_token_index": 256000,
"eos_token_id": [
1,
106
],
"image_token_index": 262144,
"initializer_range": 0.02,
"mm_tokens_per_image": 256,
"model_type": "gemma3",
"text_config": {
"_sliding_window_pattern": 6,
"attention_bias": false,
"attention_dropout": 0.0,
"attn_logit_softcapping": null,
"cache_implementation": "hybrid",
"dtype": "float32",
"final_logit_softcapping": null,
"head_dim": 16,
"hidden_activation": "gelu_pytorch_tanh",
"hidden_size": 32,
"initializer_range": 0.02,
"intermediate_size": 64,
"layer_types": [
"sliding_attention",
"full_attention"
],
"max_position_embeddings": 512,
"model_type": "gemma3_text",
"num_attention_heads": 2,
"num_hidden_layers": 2,
"num_key_value_heads": 2,
"query_pre_attn_scalar": 16,
"rms_norm_eps": 1e-06,
"rope_local_base_freq": 10000,
"rope_scaling": {
"factor": 8.0,
"rope_type": "linear"
},
"rope_theta": 1000000,
"sliding_window": 1024,
"sliding_window_pattern": 6,
"use_bidirectional_attention": false,
"use_cache": true,
"vocab_size": 262208
},
"transformers_version": "4.57.3",
"vision_config": {
"attention_dropout": 0.0,
"dtype": "float32",
"hidden_act": "gelu_pytorch_tanh",
"hidden_size": 32,
"image_size": 896,
"intermediate_size": 64,
"layer_norm_eps": 1e-06,
"model_type": "siglip_vision_model",
"num_attention_heads": 2,
"num_channels": 3,
"num_hidden_layers": 2,
"patch_size": 14,
"vision_use_head": false
}
}
tests/model_tests/diffusion/tiny_configs/LTX2Pipeline/transformer/config.json
+44
View File
@@ -0,0 +1,44 @@
{
"_class_name": "LTX2VideoTransformer3DModel",
"_diffusers_version": "0.37.0.dev0",
"activation_fn": "gelu-approximate",
"attention_bias": true,
"attention_head_dim": 16,
"attention_out_bias": true,
"audio_attention_head_dim": 16,
"audio_cross_attention_dim": 32,
"audio_hop_length": 160,
"audio_in_channels": 128,
"audio_num_attention_heads": 2,
"audio_out_channels": 128,
"audio_patch_size": 1,
"audio_patch_size_t": 1,
"audio_pos_embed_max_pos": 20,
"audio_sampling_rate": 16000,
"audio_scale_factor": 4,
"base_height": 2048,
"base_width": 2048,
"caption_channels": 32,
"causal_offset": 1,
"cross_attention_dim": 32,
"cross_attn_timestep_scale_multiplier": 1000,
"in_channels": 128,
"norm_elementwise_affine": false,
"norm_eps": 1e-06,
"num_attention_heads": 2,
"num_layers": 2,
"out_channels": 128,
"patch_size": 1,
"patch_size_t": 1,
"pos_embed_max_pos": 20,
"qk_norm": "rms_norm_across_heads",
"rope_double_precision": true,
"rope_theta": 10000.0,
"rope_type": "split",
"timestep_scale_multiplier": 1000,
"vae_scale_factors": [
8,
32,
32
]
}
tests/model_tests/diffusion/tiny_configs/LTX2Pipeline/vae/config.json
+81
View File
@@ -0,0 +1,81 @@
{
"_class_name": "AutoencoderKLLTX2Video",
"_diffusers_version": "0.37.0.dev0",
"block_out_channels": [
256,
512,
1024,
2048
],
"decoder_block_out_channels": [
256,
512,
1024
],
"decoder_causal": false,
"decoder_inject_noise": [
false,
false,
false,
false
],
"decoder_layers_per_block": [
1,
1,
1,
1
],
"decoder_spatial_padding_mode": "reflect",
"decoder_spatio_temporal_scaling": [
true,
true,
true
],
"down_block_types": [
"LTX2VideoDownBlock3D",
"LTX2VideoDownBlock3D",
"LTX2VideoDownBlock3D",
"LTX2VideoDownBlock3D"
],
"downsample_type": [
"spatial",
"temporal",
"spatiotemporal",
"spatiotemporal"
],
"encoder_causal": true,
"encoder_spatial_padding_mode": "zeros",
"in_channels": 3,
"latent_channels": 128,
"layers_per_block": [
1,
1,
1,
1,
1
],
"out_channels": 3,
"patch_size": 4,
"patch_size_t": 1,
"resnet_norm_eps": 1e-06,
"scaling_factor": 1.0,
"spatial_compression_ratio": 32,
"spatio_temporal_scaling": [
true,
true,
true,
true
],
"temporal_compression_ratio": 8,
"timestep_conditioning": false,
"upsample_factor": [
2,
2,
2
],
"upsample_residual": [
true,
true,
true
]
}