Upload folder using huggingface_hub
Browse files- config.json +99 -0
- pytorch_model.bin +3 -0
- tokenizer.json +0 -0
- tokenizer_config.json +7 -0
config.json
ADDED
|
@@ -0,0 +1,99 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"name": "parcae-xlarge-1_3b",
|
| 3 |
+
"hf_config": {
|
| 4 |
+
"org": "SandyResearch",
|
| 5 |
+
"name": "parcae-xlarge-1_3b"
|
| 6 |
+
},
|
| 7 |
+
"block_size": 2048,
|
| 8 |
+
"n_embd": 1536,
|
| 9 |
+
"intermediate_size": 6144,
|
| 10 |
+
"num_attention_heads": 12,
|
| 11 |
+
"num_key_value_heads": 12,
|
| 12 |
+
"vocab_size": 32768,
|
| 13 |
+
"padding_multiple": 64,
|
| 14 |
+
"padded_vocab_size": 32768,
|
| 15 |
+
"rope_settings": {
|
| 16 |
+
"use_rope": true,
|
| 17 |
+
"rope_condense_ratio": 1,
|
| 18 |
+
"rope_base": 50000
|
| 19 |
+
},
|
| 20 |
+
"use_abacus": false,
|
| 21 |
+
"randomize_positions_from": null,
|
| 22 |
+
"block_class_name": "TransformerPreNormBlock",
|
| 23 |
+
"norm_class_name": "RMSNorm",
|
| 24 |
+
"attention_class_name": "CausalSelfAttention",
|
| 25 |
+
"attn_impl": "flash",
|
| 26 |
+
"norm_eps": 1e-05,
|
| 27 |
+
"mlp_class_name": "BaseMLP",
|
| 28 |
+
"nonlin_name": "ReLU2",
|
| 29 |
+
"bias": false,
|
| 30 |
+
"qk_bias": false,
|
| 31 |
+
"init_strategy": "scaled-zero",
|
| 32 |
+
"init_orthogonal": true,
|
| 33 |
+
"skip_initialization": false,
|
| 34 |
+
"attention_nonlin_name": null,
|
| 35 |
+
"kv_lora_rank": null,
|
| 36 |
+
"q_lora_rank": null,
|
| 37 |
+
"qk_nope_head_dim": null,
|
| 38 |
+
"qk_rope_head_dim": null,
|
| 39 |
+
"v_head_dim": null,
|
| 40 |
+
"rve_rank": 1,
|
| 41 |
+
"rve_bias": false,
|
| 42 |
+
"rve_scale_by_sqrt_rank": false,
|
| 43 |
+
"gdn_mode": "chunk",
|
| 44 |
+
"gdn_expand_k": 0.75,
|
| 45 |
+
"gdn_expand_v": 1.5,
|
| 46 |
+
"gdn_qk_norm": "l2",
|
| 47 |
+
"gdn_conv_size": 4,
|
| 48 |
+
"gdn_conv_bias": false,
|
| 49 |
+
"gdn_gate_fn": "swish",
|
| 50 |
+
"gdn_gate_logit_normalizer": 16,
|
| 51 |
+
"gdn_fuse_norm": true,
|
| 52 |
+
"gdn_elementwise_affine": true,
|
| 53 |
+
"gdn_use_mamba_gate": true,
|
| 54 |
+
"gdn_use_mva": false,
|
| 55 |
+
"gdn_use_residual": false,
|
| 56 |
+
"gdn_use_input_gate": false,
|
| 57 |
+
"kda_mode": "chunk",
|
| 58 |
+
"kda_impl": "fla-chunk",
|
| 59 |
+
"kda_head_dim": 128,
|
| 60 |
+
"kda_num_v_heads": null,
|
| 61 |
+
"kda_expand_v": 1.0,
|
| 62 |
+
"kda_use_short_conv": true,
|
| 63 |
+
"kda_allow_neg_eigval": false,
|
| 64 |
+
"kda_safe_gate": false,
|
| 65 |
+
"kda_lower_bound": null,
|
| 66 |
+
"kda_disable_recompute": false,
|
| 67 |
+
"kda_conv_size": 4,
|
| 68 |
+
"kda_conv_bias": false,
|
| 69 |
+
"mup_model_scaling_factor": 1,
|
| 70 |
+
"use_fused_head": "liger",
|
| 71 |
+
"debias_attention": false,
|
| 72 |
+
"center_attention": false,
|
| 73 |
+
"clip_qkv": null,
|
| 74 |
+
"qk_norm": true,
|
| 75 |
+
"logit_softcap": null,
|
| 76 |
+
"activation_checkpoint_impl": "per-iteration",
|
| 77 |
+
"simple_ops": false,
|
| 78 |
+
"strategy": "single",
|
| 79 |
+
"injection_type": "diagonal",
|
| 80 |
+
"n_layers_in_recurrent_block": 8,
|
| 81 |
+
"n_layers_in_prelude": 8,
|
| 82 |
+
"n_layers_in_coda": 8,
|
| 83 |
+
"state_init": "like-init",
|
| 84 |
+
"recurrent_embedding_dimension": 1536,
|
| 85 |
+
"recurrent_intermediation_embedding_dimension": 6144,
|
| 86 |
+
"recurrent_num_attention_heads": null,
|
| 87 |
+
"prelude_norm": true,
|
| 88 |
+
"sampling_scheme": "poisson-truncated-full",
|
| 89 |
+
"recurrence_depth_variance": 12.0,
|
| 90 |
+
"mean_recurrence": 8,
|
| 91 |
+
"mean_backprop_depth": 4,
|
| 92 |
+
"lockstep_n": true,
|
| 93 |
+
"lockstep_k": false,
|
| 94 |
+
"curriculum_target": "forward",
|
| 95 |
+
"recurrent_iteration_method": "per-sequence",
|
| 96 |
+
"tie_embeddings": true,
|
| 97 |
+
"model_class_name": "Parcae",
|
| 98 |
+
"_is_recurrent_block_config": false
|
| 99 |
+
}
|
pytorch_model.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:afa28cdf67e5c3a472f422f51e47fc4ed193be054f04a605cf7858c274541e57
|
| 3 |
+
size 5355508095
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"backend": "tokenizers",
|
| 3 |
+
"is_local": true,
|
| 4 |
+
"local_files_only": false,
|
| 5 |
+
"model_max_length": 1000000000000000019884624838656,
|
| 6 |
+
"tokenizer_class": "TokenizersBackend"
|
| 7 |
+
}
|