Upload FlashSTU

README.md

@@ -0,0 +1,199 @@
+---
+library_name: transformers
+tags: []
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]

config.json

@@ -0,0 +1,27 @@
+{
+  "architectures": [
+    "FlashSTU"
+  ],
+  "auto_map": {
+    "AutoConfig": "config.FlashSTUConfig",
+    "AutoModel": "model.FlashSTU"
+  },
+  "bias": false,
+  "bsz": 4,
+  "dropout": 0.0,
+  "mlp_scale": 12,
+  "model_type": "FlashSTU",
+  "n_embd": 2304,
+  "n_heads": 9,
+  "n_layers": 7,
+  "num_eigh": 24,
+  "seq_len": 8192,
+  "softcap": 50.0,
+  "torch_dtype": "float32",
+  "transformers_version": "4.44.0",
+  "use_approx": true,
+  "use_flash_fft": true,
+  "use_hankel_L": false,
+  "vocab_size": 200064,
+  "window_size": 1024
+}

config.py

@@ -0,0 +1,40 @@
+from transformers import PretrainedConfig
+
+class FlashSTUConfig(PretrainedConfig):
+    model_type = "FlashSTU"
+
+    def __init__(
+        self,
+        bsz: int = 4,
+        n_embd: int = 2304,
+        n_heads: int = 9,
+        n_layers: int = 7,
+        seq_len: int = 8192,
+        window_size: int = 1024,
+        vocab_size: int = 200064,
+        mlp_scale: int = 12,
+        bias: bool = False,
+        dropout: float = 0.0,
+        num_eigh: int = 24,
+        use_hankel_L: bool = False,
+        use_flash_fft: bool = True,
+        use_approx: bool = True,
+        softcap: float = 50.0,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.bsz = bsz
+        self.n_embd = n_embd
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.seq_len = seq_len
+        self.window_size = window_size
+        self.vocab_size = vocab_size
+        self.mlp_scale = mlp_scale
+        self.bias = bias
+        self.dropout = dropout
+        self.num_eigh = num_eigh
+        self.use_hankel_L = use_hankel_L
+        self.use_flash_fft = use_flash_fft
+        self.use_approx = use_approx
+        self.softcap = softcap

model-00001-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7333db4c5bf7217636fe39e184d8d723061529794c7167953c4b772e8fdb2a3
+size 4563328000

model-00002-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc0f3d8a6dfa78b57f6f93b9c473505206f090dc87dcaa072b3b678906030825
+size 3377400416

model-00003-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3ecd382455c5a564824d7b9a5ad1ee430809d73f86c04ee4884188ee330eeca
+size 1843789952

model.py

@@ -0,0 +1,126 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from transformers import PreTrainedModel
+
+from stu import STU
+from modules import Attention
+from utils import get_spectral_filters, nearest_power_of_two
+from flash_stu.config import FlashSTUConfig
+
+try:
+    from flash_attn.modules.mlp import GatedMlp as MLP
+    triton_mlp = True
+except ImportError as e:
+    print(f"Unable to import Triton-based MLP: {e}. Falling back to vanilla SwiGLU MLP instead.")
+    from modules import MLP
+    triton_mlp = False
+
+try:
+    from flash_attn.ops.triton.layer_norm import RMSNorm
+except ImportError as e:
+    print(f"Unable to import Triton-based RMSNorm: {e}. Falling back to PyTorch implementation.")
+    from torch.nn import RMSNorm
+
+try:
+    from flash_attn.losses.cross_entropy import CrossEntropyLoss
+except ImportError as e:
+    print(f"Unable to import Triton-based cross entropy loss: {e}. Falling back to PyTorch implementation.")
+    from torch.nn import CrossEntropyLoss
+
+class Block(nn.Module):
+    def __init__(self, config, phi, n) -> None:
+        super(Block, self).__init__()
+        # For more complex %-split arrangements, see https://arxiv.org/pdf/2406.07887
+        self.rn_1 = RMSNorm(config.n_embd)
+        self.stu = STU(config, phi, n)
+        self.rn_2 = RMSNorm(config.n_embd)
+        self.attn = Attention(config)
+        self.rn_3 = RMSNorm(config.n_embd)
+        self.mlp = MLP(
+            config.n_embd, 
+            config.n_embd * config.mlp_scale, 
+            activation=F.silu, # Use SwiGLU
+            bias1=config.bias,
+            bias2=config.bias,
+        ) if triton_mlp else MLP(config)
+        self.rn_4 = RMSNorm(config.n_embd)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = x + self.stu(self.rn_1(x))
+        x = x + self.mlp(self.rn_2(x))
+        x = x + self.attn(self.rn_3(x))
+        x = x + self.mlp(self.rn_4(x))
+        return x
+
+class FlashSTU(PreTrainedModel):
+    config_class = FlashSTUConfig
+
+    def __init__(self, config) -> None:
+        super(FlashSTU, self).__init__(config)
+        self.config = config
+        self.n_layers = config.n_layers
+        self.n_embd = config.n_embd
+        self.mlp_scale = config.mlp_scale
+        self.seq_len = config.seq_len
+        self.n = nearest_power_of_two(self.seq_len * 2 - 1, round_up=True)
+        self.vocab_size = config.vocab_size
+        self.K = config.num_eigh
+        self.use_hankel_L = config.use_hankel_L
+        self.phi = get_spectral_filters(self.seq_len, self.K, self.use_hankel_L)
+        self.use_approx = config.use_approx
+        self.dropout = config.dropout
+        self.bias = config.bias
+        self.loss_fn = CrossEntropyLoss()
+
+        self.flash_stu = nn.ModuleDict(
+            dict(
+                tok_emb=nn.Embedding(self.vocab_size, self.n_embd),
+                dropout=nn.Dropout(self.dropout),
+                hidden=nn.ModuleList(
+                    [
+                        Block(self.config, self.phi, self.n)
+                        for _ in range(self.n_layers)
+                    ]
+                ),
+                rn_f=RMSNorm(config.n_embd)
+            )
+        )
+        self.lm_head = nn.Linear(self.n_embd, self.vocab_size, bias=self.bias)
+
+        self.std = (self.n_embd) ** -0.5
+        self.apply(self._init_weights)
+        print("Model Parameter Count: %.2fM\n" % (self._get_num_params() / 1e6,))
+
+    def forward(self, x: torch.Tensor) -> torch.tensor:
+        tok_emb = self.flash_stu.tok_emb(x)
+        x = self.flash_stu.dropout(tok_emb)
+
+        for block in self.flash_stu.hidden:
+            x = block(x)
+        x = self.flash_stu.rn_f(x)
+
+        y_hat = self.lm_head(x)
+        return y_hat
+
+    def _get_num_params(self):
+        n_params = sum(p.numel() for p in self.parameters())
+        return n_params
+
+    def _init_weights(self, module):
+        if isinstance(module, nn.Linear):
+            if hasattr(module, "SCALE_INIT"):
+                self.std *= (2 * self.n_layers) ** -0.5
+            torch.nn.init.normal_(module.weight, mean=0.0, std=self.std)
+            if module.bias is not None:
+                torch.nn.init.zeros_(module.bias)
+        elif isinstance(module, nn.Embedding):
+            torch.nn.init.normal_(module.weight, mean=0.0, std=self.std)
+        elif isinstance(module, STU):
+            if self.use_approx:
+                torch.nn.init.xavier_normal_(module.M_inputs)
+                torch.nn.init.xavier_normal_(module.M_filters)
+            else:
+                torch.nn.init.xavier_normal_(module.M_phi_plus)
+                torch.nn.init.xavier_normal_(module.M_phi_minus)

model.safetensors.index.json

@@ -0,0 +1,136 @@
+{
+  "metadata": {
+    "total_size": 9784503296
+  },
+  "weight_map": {
+    "flash_stu.hidden.0.attn.c_attn.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.attn.c_proj.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.rn_1.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.rn_2.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.rn_3.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.rn_4.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.M_filters": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.M_inputs": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.flash_fft.f_16_fft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.flash_fft.f_16_ifft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.flash_fft.f_32_fft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.flash_fft.f_32_ifft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.flash_fft.twiddle_factors_fft_16_1K": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.flash_fft.twiddle_factors_fft_32_32": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.flash_fft.twiddle_factors_ifft_16_1K": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.0.stu.flash_fft.twiddle_factors_ifft_32_32": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.attn.c_attn.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.attn.c_proj.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.rn_1.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.rn_2.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.rn_3.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.rn_4.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.M_filters": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.M_inputs": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.flash_fft.f_16_fft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.flash_fft.f_16_ifft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.flash_fft.f_32_fft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.flash_fft.f_32_ifft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.flash_fft.twiddle_factors_fft_16_1K": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.flash_fft.twiddle_factors_fft_32_32": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.flash_fft.twiddle_factors_ifft_16_1K": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.1.stu.flash_fft.twiddle_factors_ifft_32_32": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.attn.c_attn.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.attn.c_proj.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.mlp.fc1.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.mlp.fc2.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.rn_1.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.rn_2.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.rn_3.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.rn_4.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.M_filters": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.M_inputs": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.flash_fft.f_16_fft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.flash_fft.f_16_ifft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.flash_fft.f_32_fft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.flash_fft.f_32_ifft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.flash_fft.twiddle_factors_fft_16_1K": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.flash_fft.twiddle_factors_fft_32_32": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.flash_fft.twiddle_factors_ifft_16_1K": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.2.stu.flash_fft.twiddle_factors_ifft_32_32": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.attn.c_attn.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.attn.c_proj.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.3.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.3.rn_1.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.rn_2.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.rn_3.weight": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.rn_4.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.M_filters": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.M_inputs": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.flash_fft.f_16_fft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.flash_fft.f_16_ifft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.flash_fft.f_32_fft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.flash_fft.f_32_ifft": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.flash_fft.twiddle_factors_fft_16_1K": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.flash_fft.twiddle_factors_fft_32_32": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.flash_fft.twiddle_factors_ifft_16_1K": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.3.stu.flash_fft.twiddle_factors_ifft_32_32": "model-00001-of-00003.safetensors",
+    "flash_stu.hidden.4.attn.c_attn.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.attn.c_proj.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.rn_1.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.rn_2.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.rn_3.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.rn_4.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.M_filters": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.M_inputs": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.flash_fft.f_16_fft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.flash_fft.f_16_ifft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.flash_fft.f_32_fft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.flash_fft.f_32_ifft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.flash_fft.twiddle_factors_fft_16_1K": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.flash_fft.twiddle_factors_fft_32_32": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.flash_fft.twiddle_factors_ifft_16_1K": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.4.stu.flash_fft.twiddle_factors_ifft_32_32": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.attn.c_attn.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.attn.c_proj.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.rn_1.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.rn_2.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.rn_3.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.rn_4.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.M_filters": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.M_inputs": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.flash_fft.f_16_fft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.flash_fft.f_16_ifft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.flash_fft.f_32_fft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.flash_fft.f_32_ifft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.flash_fft.twiddle_factors_fft_16_1K": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.flash_fft.twiddle_factors_fft_32_32": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.flash_fft.twiddle_factors_ifft_16_1K": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.5.stu.flash_fft.twiddle_factors_ifft_32_32": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.attn.c_attn.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.attn.c_proj.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.mlp.fc1.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.mlp.fc2.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.rn_1.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.rn_2.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.rn_3.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.rn_4.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.M_filters": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.M_inputs": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.flash_fft.f_16_fft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.flash_fft.f_16_ifft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.flash_fft.f_32_fft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.flash_fft.f_32_ifft": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.flash_fft.twiddle_factors_fft_16_1K": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.flash_fft.twiddle_factors_fft_32_32": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.flash_fft.twiddle_factors_ifft_16_1K": "model-00002-of-00003.safetensors",
+    "flash_stu.hidden.6.stu.flash_fft.twiddle_factors_ifft_32_32": "model-00002-of-00003.safetensors",
+    "flash_stu.rn_f.weight": "model-00002-of-00003.safetensors",
+    "flash_stu.tok_emb.weight": "model-00001-of-00003.safetensors",
+    "lm_head.weight": "model-00003-of-00003.safetensors"
+  }
+}