updated config and weights

README.md

@@ -5,7 +5,7 @@ pipeline_tag: time-series-forecasting
 tags:
 - transformers
 - timesfm
-- timesfm_2p5
+- timesfm2_5
 - time-series-forecasting
 - arxiv:2310.10688
 ---
@@ -16,13 +16,12 @@ TimesFM (Time Series Foundation Model) is a pretrained decoder-only model for ti
 
 **Resources and Technical Documentation**:
 * Original model: [google/timesfm-2.5-200m-pytorch](https://huggingface.co/google/timesfm-2.5-200m-pytorch)
-* Transformers model: [google/timesfm-2.5-200m-transformers](https://huggingface.co/google/timesfm-2.5-200m-transformers)
 * Paper: [A decoder-only foundation model for time-series forecasting](https://huggingface.co/papers/2310.10688)
 * Transformers docs: [TimesFM 2.5](https://huggingface.co/docs/transformers/main/en/model_doc/timesfm_2p5)
 
 ## Model description
 
-This model is converted from the official TimesFM 2.5 PyTorch checkpoint and integrated into `transformers` as `Timesfm2P5ModelForPrediction`.
+This model is converted from the official TimesFM 2.5 PyTorch checkpoint and integrated into `transformers` as `TimesFm2_5ModelForPrediction`.
 
 The converted checkpoint preserves the original architecture and forecasting behavior, including:
 * patch-based inputs for time-series contexts
@@ -35,7 +34,7 @@ The converted checkpoint preserves the original architecture and forecasting beh
 import torch
 from transformers import Timesfm2P5ModelForPrediction
 
-model = Timesfm2P5ModelForPrediction.from_pretrained("google/timesfm-2.5-200m-transformers", attn_implementation="sdpa")
+model = TimesFm2_5ModelForPrediction.from_pretrained("google/timesfm-2.5-200m-transformers")
 model = model.to(torch.float32).eval()
 
 past_values = [

config.json

@@ -1,54 +1,26 @@
 {
   "activation": "swish",
   "architectures": [
-    "Timesfm2P5ModelForPrediction"
+    "TimesFm2_5ModelForPrediction"
   ],
   "attention_bias": false,
   "attention_dropout": 0.0,
-  "attn_logit_softcapping": null,
   "context_length": 16384,
   "decode_index": 5,
   "dtype": "float32",
   "force_flip_invariance": true,
-  "freq_size": 10,
   "head_dim": 80,
   "hidden_size": 1280,
   "horizon_length": 128,
   "infer_is_positive": true,
   "initializer_range": 0.02,
   "intermediate_size": 1280,
-  "layer_types": [
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention",
-    "attention"
-  ],
   "max_position_embeddings": 16384,
-  "max_timescale": 10000.0,
-  "min_timescale": 1.0,
-  "model_type": "timesfm_2p5",
-  "normalize_inputs": true,
+  "model_type": "timesfm2_5",
   "num_attention_heads": 16,
   "num_hidden_layers": 20,
   "num_key_value_heads": 16,
   "output_quantile_len": 1024,
-  "pad_val": -1000000000.0,
   "patch_length": 32,
   "quantiles": [
     0.1,
@@ -61,20 +33,12 @@
     0.8,
     0.9
   ],
-  "query_pre_attn_scalar": 256.0,
   "rms_norm_eps": 1e-06,
   "rope_parameters": {
     "rope_theta": 10000.0,
     "rope_type": "default"
   },
-  "rope_theta": 10000.0,
-  "sliding_window": null,
-  "tolerance": 1e-05,
   "transformers_version": "5.3.0.dev0",
   "use_bias": false,
-  "use_continuous_quantile_head": true,
-  "use_per_dim_scale": true,
-  "use_positional_embedding": false,
-  "use_qk_norm": true,
-  "use_rotary_embeddings": true
+  "use_continuous_quantile_head": true
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3ecaed5ac5d6a4ab3681fde3647fa6b5e47614f168b25ca2313d3bd7500c3fd
-size 944863880
+oid sha256:b53f6d52114e2ad786890f3c4637ce05f580b7800d6e24401f88b398b76035ef
+size 925187448