Initial anonymous commit

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 +1,19 @@
+{
+  "architectures": [
+    "PRISM_WikiText_Model"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_prism.PRISMConfig",
+    "AutoModel": "modeling_prism.PRISM_WikiText_Model"
+  },
+  "d_model": 512,
+  "dropout": 0.1,
+  "dtype": "float32",
+  "fft_dim": 64,
+  "model_type": "prism-v2",
+  "prism_depth": 5,
+  "seq_len": 4096,
+  "trans_depth": 1,
+  "transformers_version": "4.57.3",
+  "vocab_size": 32768
+}

configuration_prism.py

@@ -0,0 +1,24 @@
+
+from transformers import PretrainedConfig
+
+class PRISMConfig(PretrainedConfig):
+    model_type = "prism-v2"
+    def __init__(
+        self,
+        vocab_size=32768,
+        d_model=512,
+        seq_len=4096,
+        prism_depth=5,
+        trans_depth=1,
+        fft_dim=64,
+        dropout=0.1,
+        **kwargs
+    ):
+        super().__init__(**kwargs)
+        self.vocab_size = vocab_size
+        self.d_model = d_model
+        self.seq_len = seq_len
+        self.prism_depth = prism_depth
+        self.trans_depth = trans_depth
+        self.fft_dim = fft_dim
+        self.dropout = dropout

modeling_prism.py

@@ -0,0 +1,220 @@
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import math
+from transformers import PreTrainedModel
+
+try:
+    from .configuration_prism import PRISMConfig
+except ImportError:
+    from configuration_prism import PRISMConfig
+
+try:
+    from x_transformers import Encoder
+except ImportError:
+    raise ImportError("To use PRISM, you must run: pip install x-transformers")
+
+# --- UTILS ---
+class ComplexDropout(nn.Module):
+    def __init__(self, p=0.5):
+        super().__init__()
+        self.p = p
+    def forward(self, z):
+        if not self.training or self.p == 0.0: return z
+        mask = torch.ones_like(z.real)
+        mask = F.dropout(mask, self.p, self.training, inplace=False)
+        return z * mask
+
+class RobustPhaseNorm(nn.Module):
+    def __init__(self, d_model, eps=1e-5):
+        super().__init__()
+        self.scale = nn.Parameter(torch.ones(d_model))
+        self.eps = eps
+    def forward(self, x):
+        mag = torch.abs(x)
+        rms = torch.sqrt(torch.mean(mag**2, dim=-1, keepdim=True) + self.eps)
+        return (x / rms) * self.scale
+
+class ModReLU(nn.Module):
+    def __init__(self, features):
+        super().__init__()
+        self.b = nn.Parameter(torch.zeros(features))
+    def forward(self, z):
+        mag = torch.abs(z)
+        new_mag = F.relu(mag + self.b)
+        phase = z / (mag + 1e-6)
+        return new_mag * phase
+
+class ComplexToRealBridge(nn.Module):
+    def __init__(self, d_model):
+        super().__init__()
+        self.proj = nn.Linear(d_model * 2, d_model)
+        self.norm = nn.LayerNorm(d_model)
+    def forward(self, x_complex):
+        cat = torch.cat([x_complex.real, x_complex.imag], dim=-1)
+        return self.norm(self.proj(cat))
+
+# --- COMPONENTS ---
+class DynamicRoSE(nn.Module):
+    def __init__(self, num_embeddings, embedding_dim, max_period=10000.0):
+        super().__init__()
+        self.embedding_dim = embedding_dim
+        self.raw_embedding = nn.Embedding(num_embeddings, embedding_dim)
+        self.adapter = nn.Linear(embedding_dim, embedding_dim * 2)
+        freqs = torch.exp(torch.arange(0, embedding_dim, dtype=torch.float32) * -(math.log(max_period) / embedding_dim))
+        self.register_buffer('freqs', freqs)
+        self.rotation_predictor = nn.Linear(embedding_dim, embedding_dim * 2)
+
+    def forward(self, input_ids):
+        real_base = self.raw_embedding(input_ids)
+        B, L, D = real_base.shape
+        complex_params = self.adapter(real_base)
+        z_t = torch.complex(complex_params[..., :D], complex_params[..., D:])
+        rot_raw = self.rotation_predictor(real_base)
+        rot_x, rot_y = rot_raw.chunk(2, dim=-1)
+        rot_mag = torch.sqrt(rot_x**2 + rot_y**2 + 1e-6)
+        dynamic_rot = torch.complex(rot_x / rot_mag, rot_y / rot_mag)
+        pos = torch.arange(L, device=input_ids.device).float()
+        static_angles = torch.outer(pos, self.freqs)
+        static_rot = torch.polar(torch.ones_like(static_angles), static_angles)
+        z_final = z_t * static_rot.unsqueeze(0) * dynamic_rot
+        return z_final, real_base
+
+class HyenaNeuralFilter(nn.Module):
+    def __init__(self, d_model, max_len=1024, hidden_dim=64):
+        super().__init__()
+        self.d_model = d_model
+        freqs = torch.exp(torch.arange(0, hidden_dim, 2, dtype=torch.float32) * -(math.log(10000.0) / hidden_dim))
+        self.register_buffer("freqs", freqs)
+        self.mlp = nn.Sequential(
+            nn.Linear(hidden_dim, hidden_dim), nn.SiLU(),
+            nn.Linear(hidden_dim, hidden_dim), nn.SiLU(),
+            nn.Linear(hidden_dim, d_model * 2)
+        )
+    def forward(self, L, device):
+        t = torch.linspace(0, 1, steps=L, device=device).unsqueeze(-1)
+        emb = torch.cat([torch.sin(t * self.freqs), torch.cos(t * self.freqs)], dim=-1)
+        out = self.mlp(emb).view(L, self.d_model, 2)
+        return torch.complex(out[..., 0], out[..., 1])
+
+class GatedHarmonicConvolution(nn.Module):
+    def __init__(self, d_model, max_len=1024, dropout=0.1, hidden_dim=64):
+        super().__init__()
+        self.d_model = d_model
+        self.filter_len = max_len
+        self.neural_filter = HyenaNeuralFilter(d_model, max_len=max_len, hidden_dim=hidden_dim)
+        self.gate_proj = nn.Linear(d_model * 2, d_model * 2)
+        self.mix_real = nn.Linear(d_model, d_model)
+        self.mix_imag = nn.Linear(d_model, d_model)
+        self.out_real = nn.Linear(d_model, d_model)
+        self.out_imag = nn.Linear(d_model, d_model)
+        self.activation = ModReLU(d_model)
+        self.norm = RobustPhaseNorm(d_model)
+        self.dropout = ComplexDropout(dropout)
+
+    def forward(self, x, src_mask=None):
+        residual = x
+        x_norm = self.norm(x)
+        if src_mask is not None:
+             x_norm = x_norm.masked_fill(src_mask.unsqueeze(-1), 0.0)
+        B, L, D = x_norm.shape
+        eff_L = min(L, self.filter_len)
+        x_freq = torch.fft.fft(x_norm, n=eff_L, dim=1, norm='ortho')
+        h = self.neural_filter(eff_L, x.device).unsqueeze(0)
+        x_filtered = x_freq * h
+        x_time = torch.fft.ifft(x_filtered, n=eff_L, dim=1, norm='ortho')
+        if L > eff_L: x_time = F.pad(x_time, (0,0,0,L-eff_L))
+        else: x_time = x_time[:, :L, :]
+        gates = torch.sigmoid(self.gate_proj(torch.cat([x_norm.real, x_norm.imag], dim=-1)))
+        g_r, g_i = gates.chunk(2, dim=-1)
+        x_gated = torch.complex(x_time.real * g_r, x_time.imag * g_i)
+        mr, mi = self.mix_real, self.mix_imag
+        x_mixed = torch.complex(mr(x_gated.real) - mi(x_gated.imag), mr(x_gated.imag) + mi(x_gated.real))
+        x_act = self.activation(x_mixed)
+        or_, oi = self.out_real, self.out_imag
+        out = torch.complex(or_(x_act.real) - oi(x_act.imag), or_(x_act.imag) + oi(x_act.real))
+        return self.dropout(out) + residual
+
+class PRISMEncoder(nn.Module):
+    def __init__(self, num_layers, d_model, max_len, dropout=0.1, hidden_dim=64):
+        super().__init__()
+        self.layers = nn.ModuleList([
+            GatedHarmonicConvolution(d_model, max_len, dropout, hidden_dim)
+            for _ in range(num_layers)
+        ])
+        self.final_norm = RobustPhaseNorm(d_model)
+    def forward(self, x, src_mask=None):
+        for layer in self.layers:
+            if self.training: x = torch.utils.checkpoint.checkpoint(layer, x, src_mask, use_reentrant=False)
+            else: x = layer(x, src_mask)
+        return self.final_norm(x)
+
+# --- MAIN MODEL ---
+class PRISM_WikiText_Model(PreTrainedModel):
+    config_class = PRISMConfig
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.config = config
+        
+        self.d_model = config.d_model
+        
+        # 1. PRISM Core
+        self.rose = DynamicRoSE(config.vocab_size, config.d_model)
+        self.prism_encoder = PRISMEncoder(
+            config.prism_depth, 
+            config.d_model, 
+            max_len=config.seq_len, 
+            dropout=config.dropout,
+            hidden_dim=config.fft_dim
+        )
+        self.bridge = ComplexToRealBridge(config.d_model)
+        self.periscope_proj = nn.Sequential(
+            nn.Linear(config.d_model * 2, config.d_model), 
+            nn.LayerNorm(config.d_model), 
+            nn.GELU()
+        )
+
+        # 2. Refiner (Standard Transformer + RoPE)
+        if config.trans_depth > 0:
+            self.refiner = Encoder(
+                dim=config.d_model,
+                depth=config.trans_depth,
+                heads=8,
+                rotary_pos_emb=True,
+                attn_flash=True,
+                attn_dropout=config.dropout,
+                ff_dropout=config.dropout,
+            )
+        else:
+            self.refiner = None
+
+        # 3. Output
+        self.lm_head = nn.Linear(config.d_model, config.vocab_size)
+        self.lm_head.weight = self.rose.raw_embedding.weight
+
+    def forward(self, input_ids, labels=None):
+        # A. Wave Physics
+        wave_src, particle_src = self.rose(input_ids)
+        wave_out = self.prism_encoder(wave_src)
+        wave_real = self.bridge(wave_out)
+
+        # B. Interface
+        mixed_memory = self.periscope_proj(torch.cat([wave_real, particle_src], dim=-1))
+
+        # C. Digital Refinement
+        if self.refiner:
+            out = self.refiner(mixed_memory)
+        else:
+            out = mixed_memory
+
+        logits = self.lm_head(out)
+        
+        loss = None
+        if labels is not None:
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(logits.view(-1, self.config.vocab_size), labels.view(-1))
+            return {"loss": loss, "logits": logits}
+
+        return logits

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ac47f33b73d047408abba88efa05737ddf9bc8fb8154f0f51536ea8fd4a8169
+size 131815503