Overhaul: REAP-CRACK v2 — fresh Q4 surgery on REAP base, 37 tok/s, 5/5 compliance

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
+proof_napalm.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,110 +1,85 @@
 ---
 license: other
-license_name: gated-research
-license_link: LICENSE
+license_name: tongyi-qianwen
+license_link: https://huggingface.co/Qwen/Qwen3-235B-A22B/blob/main/LICENSE
+base_model:
+- mlx-community/Qwen3.5-397B-A17B-4bit
 language:
 - en
-- zh
 tags:
-- abliteration
-- safety-research
 - mlx
-- qwen
+- abliterated
+- uncensored
+- qwen3
 - moe
-- reap
-- crack
-library_name: mlx
-pipeline_tag: text-generation
-base_model: Qwen/Qwen3.5-397B-A17B
-extra_gated_heading: "Access Request Required"
-extra_gated_description: "This is a gated research model with modified safety behaviors. Please describe your intended use case to request access."
-extra_gated_fields:
-  Name: text
-  Organization: text
-  Country: text
-  Intended Use: text
-  I agree to use this model for research purposes only: checkbox
-  I understand this model has been abliterated and will not use it for harm: checkbox
 ---
 
-# Qwen3.5-397B-A17B-REAP-CRACK
+# Qwen 3.5 397B-A17B — REAP-CRACK (4-bit MLX)
 
-<p align="center">
-  <img src="logo.png" alt="Dealign.ai" width="180"/>
-  <br/>
-  <strong><a href="https://dealign.ai">Dealign.ai</a></strong>
-</p>
+> **Abliterated** variant of Qwen 3.5 397B MoE with permanent refusal removal via weight surgery.
 
-**Abliterated and expert-pruned Qwen 3.5 397B — optimized for Apple Silicon.**
+## What Is This?
 
-> ⚠️ **Gated Research Model** — This model has had safety guardrails significantly reduced. Access is granted on a case-by-case basis for research purposes.
+This is [Qwen 3.5 397B-A17B](https://huggingface.co/Qwen/Qwen3-235B-A22B) (4-bit quantized for MLX) with **permanent abliteration** — the model's refusal behavior has been surgically removed at the weight level. No custom model files, no runtime hooks, no steering vectors. Just a standard MLX model that runs at full speed.
 
-## Overview
+### Key Specs
 
-This is a **REAP-pruned** and **CRACK-abliterated** version of Qwen 3.5 397B-A17B, the hybrid GatedDeltaNet/Full-Attention MoE model from Qwen. 
+| Metric | Value |
+|--------|-------|
+| **Architecture** | Qwen 3.5 MoE (397B total, 17B active) |
+| **Quantization** | 4-bit, group_size=64, affine mode |
+| **Speed** | ~37 tok/s on Mac Studio M2 Ultra (256GB) |
+| **Surgery Layers** | L27 + L31 `self_attn.o_proj` (full attention layers) |
+| **Surgery Strength** | s=10 (fresh Q4 quantization) |
+| **Custom model.py** | ❌ None needed — uses built-in `qwen3_5.py` |
 
-The abliteration was performed using a novel architecture-aware technique developed through **71 controlled experiments** that account for the unique information pathways in hybrid SSM/attention architectures. Unlike standard abliteration methods that fail on this architecture, our approach achieves high compliance with **improved** language modeling quality — perplexity actually decreased after abliteration.
+## Proof It Works
 
-REAP (Redundant Expert Automated Pruning) reduces the expert count from 512 to 397 per layer, cutting model size while preserving quality.
+![Proof: Model answering about napalm at 37.2 tok/s in vMLX](proof_napalm.png)
 
-## Performance Highlights
+*1166 tokens at 37.2 t/s — full compliance with no refusal, running natively in vMLX on Mac Studio.*
 
-- **High compliance** across safety categories with minimal refusals
-- **Improved perplexity** — abliteration removed a noise-adding direction, making the model *better* at language modeling
-- **Expert-pruned** — 22% fewer experts, smaller memory footprint
-- **Drop-in MLX usage** — loads with standard `mlx_lm.load()`, no additional setup
+## How It Was Made
 
-## Requirements
+This model uses **CRACK** (Controlled Refusal Ablation via Calibrated Knockouts) — a research tool for removing refusal behavior from quantized LLMs.
 
-- **Hardware**: Apple Silicon Mac with ≥192GB unified memory (256GB recommended)
-- **Framework**: MLX (`pip install mlx-lm`)
-- **Disk**: ~171GB
+### Technical Details
+
+1. **Refusal vector extraction** at Layer 28 (post-SSM, where refusal signal consolidates in Qwen 3.5's hybrid GatedDeltaNet architecture)
+2. **Weight surgery**: `W' = W - s × v @ (vᵀ @ W)` applied to `o_proj` at L27 + L31 (full attention layers — no SSM bypass channel)
+3. **Fresh Q4 quantization**: Surgery performed on FP16 weights, then re-quantized to Q4 with `mx.quantize()` computing new optimal scales/biases
+4. **Binary shard patching**: Modified tensor data injected directly into original safetensors binary format, preserving all metadata, tensor ordering, and bf16 dtypes for maximum inference speed
+
+### Why These Specific Layers?
+
+Qwen 3.5 uses a **hybrid SSM/attention** architecture. Every 4th layer is full attention; the rest are GatedDeltaNet (SSM). Refusal signal can bypass residual-stream interventions via the SSM recurrent state. L27 and L31 are full attention layers that bracket the critical L28 refusal consolidation point — surgery here cannot be routed around.
 
 ## Usage
 
+### With mlx-lm
 ```python
 from mlx_lm import load, generate
 
-model, tok = load("dealignai/Qwen3.5-397B-A17B-REAP-CRACK")
-
-prompt = tok.apply_chat_template(
+model, tokenizer = load("dealignai/Qwen3.5-397B-A17B-REAP-CRACK")
+prompt = tokenizer.apply_chat_template(
     [{"role": "user", "content": "Your prompt here"}],
     add_generation_prompt=True, tokenize=False, enable_thinking=False
 )
-response = generate(model, tok, prompt=prompt, max_tokens=512)
+response = generate(model, tokenizer, prompt=prompt, max_tokens=500)
 print(response)
 ```
 
-## Model Details
-
-| Property | Value |
-|----------|-------|
-| Base Model | Qwen/Qwen3.5-397B-A17B |
-| Parameters | 397B total, 17B active per forward pass |
-| Architecture | Hybrid GatedDeltaNet + Full Attention MoE |
-| Experts | 397/layer (REAP-pruned from 512) |
-| Quantization | 4-bit mixed precision |
-| Model Size | ~171GB |
-| Abliteration | Architecture-aware weight surgery (CRACK) |
-
-## About the Technique
-
-The abliteration leverages architectural properties unique to hybrid SSM/attention models. Standard abliteration techniques (CAA, orthogonal projection, etc.) are ineffective on this architecture due to the presence of multiple information pathways that can bypass residual-stream interventions.
-
-Our technique — developed over 71 experiments across 16 intervention paradigms — identifies and exploits the specific layers and weight matrices where the safety signal is most vulnerable, while avoiding the quantization noise floor that prevents standard weight surgery at 4-bit precision.
-
-Full research details are maintained privately.
-
-## About Dealign.ai
+### With vMLX
+Point vMLX to this model directory. It will auto-detect as `qwen3_5_moe` and load via the optimized built-in path.
 
-[Dealign.ai](https://dealign.ai) researches alignment mechanisms in large language models. Our tools include **CRACK** (abliteration engine) and **REAP** (expert pruning), designed for understanding and modifying safety behaviors in frontier models.
+## Base Model
 
-## Ethical Statement
+Based on [mlx-community/Qwen3.5-397B-A17B-4bit](https://huggingface.co/mlx-community/Qwen3.5-397B-A17B-4bit) with expert pruning (REAP — Routing-Efficient Adaptive Pruning).
 
-This model is released under **gated access for alignment research purposes only**. Understanding how safety mechanisms work — and where they fail — in modern architectures is essential for building more robust AI systems. 
+## Research
 
-The creators are not responsible for any misuse. By requesting access, you agree to use this model responsibly and in compliance with applicable laws and regulations.
+This model is part of ongoing research into alignment removal techniques for large language models. See the [CRACK project](https://github.com/exploitbot/CRACK_abliteration) for details.
 
-## Disclaimer
+## ⚠️ Disclaimer
 
-THIS MODEL IS PROVIDED "AS-IS" FOR RESEARCH PURPOSES. NO WARRANTY IS PROVIDED. USE AT YOUR OWN RISK.
+This model has had safety guardrails removed. It will comply with requests that the base model would refuse. Use responsibly and in accordance with applicable laws.

config.json

@@ -2,8 +2,22 @@
   "architectures": [
     "Qwen3_5MoeForConditionalGeneration"
   ],
+  "eos_token_id": [
+    248046,
+    248044
+  ],
   "image_token_id": 248056,
   "model_type": "qwen3_5_moe",
+  "quantization": {
+    "group_size": 64,
+    "bits": 4,
+    "mode": "affine"
+  },
+  "quantization_config": {
+    "group_size": 64,
+    "bits": 4,
+    "mode": "affine"
+  },
   "text_config": {
     "attention_bias": false,
     "attention_dropout": 0.0,
@@ -114,43 +128,6 @@
   "tie_word_embeddings": false,
   "transformers_version": "4.57.0.dev0",
   "video_token_id": 248057,
-  "vision_config": {
-    "deepstack_visual_indexes": [],
-    "depth": 27,
-    "hidden_act": "gelu_pytorch_tanh",
-    "hidden_size": 1152,
-    "in_channels": 3,
-    "initializer_range": 0.02,
-    "intermediate_size": 4304,
-    "model_type": "qwen3_5_moe",
-    "num_heads": 16,
-    "num_position_embeddings": 2304,
-    "out_hidden_size": 4096,
-    "patch_size": 16,
-    "spatial_merge_size": 2,
-    "temporal_patch_size": 2
-  },
   "vision_end_token_id": 248054,
-  "vision_start_token_id": 248053,
-  "quantization": {
-    "group_size": 64,
-    "bits": 4,
-    "mode": "affine"
-  },
-  "quantization_config": {
-    "group_size": 64,
-    "bits": 4,
-    "mode": "affine"
-  },
-  "model_file": "model.py",
-  "crack_steering": {
-    "layers": [
-      28,
-      29
-    ],
-    "alpha": 4.0,
-    "vector_file": "steering_direction.safetensors",
-    "method": "contrastive_activation_addition",
-    "note": "Built-in CAA steering \u2014 no external code needed"
-  }
+  "vision_start_token_id": 248053
 }

model.py

@@ -1,331 +0,0 @@
-# Copyright © 2026 Apple Inc. (base model)
-# CRACK-REAP Steering Extension by CRACK Research Team
-#
-# This is a custom model.py that extends Qwen3_5MoeForConditionalGeneration
-# with FP16 weight surgery at L27+L31 o_proj (Finding #70).
-#
-# Ablated FP16 o_proj weights are loaded from 'surgery/ablated_oproj.safetensors'
-# in the model directory. Surgery removes the refusal direction permanently
-# from the attention output projection weights. No runtime steering needed.
-#
-# Usage: Just load normally with mlx_lm.load("/path/to/model/")
-# The steering is applied automatically during inference.
-
-from dataclasses import dataclass, field
-from typing import Any, Dict, List, Optional, Tuple, Union
-
-import mlx.core as mx
-import mlx.nn as nn
-import numpy as np
-from mlx.utils import tree_flatten, tree_unflatten
-from pathlib import Path
-import os
-
-from mlx_lm.models.base import (
-    BaseModelArgs,
-    create_attention_mask,
-    create_ssm_mask,
-)
-from mlx_lm.models.cache import ArraysCache, KVCache
-from mlx_lm.models.qwen3_next import Qwen3NextAttention as Attention
-from mlx_lm.models.qwen3_next import Qwen3NextMLP as MLP
-from mlx_lm.models.qwen3_next import Qwen3NextRMSNormGated as RMSNormGated
-from mlx_lm.models.qwen3_next import Qwen3NextSparseMoeBlock as SparseMoeBlock
-
-# =====================================================================
-# Steering Configuration — baked into the model
-# =====================================================================
-SURGERY_LAYERS = [27, 31]   # Layers with FP16 ablated o_proj
-SURGERY_STRENGTH = 5         # 5=quality (10/12, PPL-12), 10=max (11/12, PPL-14)
-STEERING_ALPHA = 0.0         # Disabled — surgery handles abliteration permanently
-
-
-# =====================================================================
-# Model Args (identical to base qwen3_5_moe)
-# =====================================================================
-
-@dataclass
-class TextModelArgs(BaseModelArgs):
-    model_type: str = ""
-    hidden_size: int = 4096
-    intermediate_size: int = 14336
-    num_hidden_layers: int = 32
-    num_attention_heads: int = 32
-    rms_norm_eps: float = 1e-6
-    vocab_size: int = 151936
-    num_key_value_heads: int = 8
-    max_position_embeddings: int = 131072
-    linear_num_value_heads: int = 64
-    linear_num_key_heads: int = 16
-    linear_key_head_dim: int = 192
-    linear_value_head_dim: int = 128
-    linear_conv_kernel_dim: int = 4
-    tie_word_embeddings: bool = False
-    attention_bias: bool = False
-    head_dim: Optional[int] = None
-    full_attention_interval: int = 4
-
-    num_experts: int = 0
-    num_experts_per_tok: int = 0
-    decoder_sparse_step: int = 1
-    shared_expert_intermediate_size: int = 0
-    moe_intermediate_size: int = 0
-    norm_topk_prob: bool = True
-
-    rope_parameters: Optional[Dict[str, Union[float, str, bool, List[int]]]] = field(
-        default_factory=lambda: {
-            "type": "default",
-            "mrope_section": [11, 11, 10],
-            "rope_theta": 100000,
-            "partial_rotary_factor": 0.25,
-        }
-    )
-
-    partial_rotary_factor: float = 0.25
-    rope_theta: float = 100000.0
-    rope_scaling: Optional[Dict[str, Union[float, str]]] = None
-
-    def __post_init__(self):
-        if self.head_dim is None:
-            self.head_dim = self.hidden_size // self.num_attention_heads
-        if self.rope_parameters:
-            if "type" not in self.rope_parameters and "rope_type" in self.rope_parameters:
-                self.rope_parameters["type"] = self.rope_parameters.pop("rope_type")
-            self.partial_rotary_factor = self.rope_parameters.get("partial_rotary_factor", 0.25)
-            self.rope_theta = self.rope_parameters.get("rope_theta", 100000.0)
-            self.rope_scaling = self.rope_parameters
-
-
-# =====================================================================
-# Import remaining components from qwen3_5 (unchanged)
-# =====================================================================
-from mlx_lm.models.qwen3_5 import (
-    GatedDeltaNet,
-    DecoderLayer,
-)
-
-
-# =====================================================================
-# Steered Text Model — the key change
-# =====================================================================
-
-class SteeredQwen3_5TextModel(nn.Module):
-    """Qwen3_5TextModel with built-in CAA steering at layers 28-29."""
-
-    def __init__(self, args: TextModelArgs):
-        super().__init__()
-        self.embed_tokens = nn.Embedding(args.vocab_size, args.hidden_size)
-        self.layers = [
-            DecoderLayer(args=args, layer_idx=i)
-            for i in range(args.num_hidden_layers)
-        ]
-        self.norm = nn.RMSNorm(args.hidden_size, eps=args.rms_norm_eps)
-        self.ssm_idx = 0
-        self.fa_idx = args.full_attention_interval - 1
-
-        # Steering direction — loaded from steering_direction.safetensors
-        # Shape: [1, 1, hidden_size] for broadcasting
-        self.steering_direction = mx.zeros((1, 1, args.hidden_size))
-        self.steering_alpha = STEERING_ALPHA
-        self.steering_layers = set(SURGERY_LAYERS)
-
-        # FP16 Surgery: load ablated o_proj weights and replace QuantizedLinear
-        self._surgery_loaded = False
-
-    def _load_surgery(self):
-        """Load FP16 ablated o_proj weights and replace QuantizedLinear layers."""
-        if self._surgery_loaded:
-            return
-        import glob
-        # Find model directory from the embed_tokens weight location
-        # Surgery file is co-located with safetensors shards
-        model_dir = None
-        for p in ["/Volumes/EricsLLMDrive/Qwen3.5-397B-A17B-CRACK-V5-4bit",
-                   str(Path(__file__).parent)]:
-            sf = os.path.join(p, "surgery/ablated_oproj.safetensors")
-            if os.path.exists(sf):
-                model_dir = p
-                break
-        if model_dir is None:
-            print("[CRACK] No surgery file found, using activation steering fallback")
-            self.steering_alpha = 4.1  # Fallback to steering
-            self._surgery_loaded = True
-            return
-        surgery_path = os.path.join(model_dir, "surgery/ablated_oproj.safetensors")
-        surgery_w = mx.load(surgery_path)
-        s_prefix = f"s{SURGERY_STRENGTH}"
-        for layer_idx in SURGERY_LAYERS:
-            key = f"{s_prefix}.layers.{layer_idx}.self_attn.o_proj.weight"
-            if key not in surgery_w:
-                print(f"[CRACK] WARNING: {key} not in surgery file")
-                continue
-            w_fp16 = surgery_w[key]
-            new_linear = nn.Linear(w_fp16.shape[1], w_fp16.shape[0], bias=False)
-            new_linear.weight = w_fp16
-            self.layers[layer_idx].self_attn.o_proj = new_linear
-            print(f"[CRACK] L{layer_idx} o_proj: Q4 -> FP16 surgery (s={SURGERY_STRENGTH})")
-        self._surgery_loaded = True
-
-    def __call__(
-        self,
-        inputs: mx.array,
-        cache: Optional[Any] = None,
-        input_embeddings: Optional[mx.array] = None,
-    ) -> mx.array:
-        # Auto-load surgery on first call (after model weights are loaded)
-        if not self._surgery_loaded:
-            self._load_surgery()
-
-        if input_embeddings is not None:
-            hidden_states = input_embeddings
-        else:
-            hidden_states = self.embed_tokens(inputs)
-
-        if cache is None:
-            cache = [None] * len(self.layers)
-
-        fa_mask = create_attention_mask(hidden_states, cache[self.fa_idx])
-        ssm_mask = create_ssm_mask(hidden_states, cache[self.ssm_idx])
-
-        for i, (layer, c) in enumerate(zip(self.layers, cache)):
-            mask = ssm_mask if layer.is_linear else fa_mask
-            hidden_states = layer(hidden_states, mask=mask, cache=c)
-
-            # ============ CAA STEERING ============
-            # Apply contrastive activation steering after target layers.
-            # This is the proven 6/6 compliance operation:
-            #   h_new = h - α * (v^T · h) * v
-            # Applied at every token, every generation step.
-            if i in self.steering_layers:
-                proj = mx.sum(
-                    hidden_states * self.steering_direction,
-                    axis=-1, keepdims=True
-                )
-                hidden_states = hidden_states - self.steering_alpha * proj * self.steering_direction
-            # ======================================
-
-        return self.norm(hidden_states)
-
-
-class TextModel(nn.Module):
-    def __init__(self, args: TextModelArgs):
-        super().__init__()
-        self.args = args
-        self.model_type = args.model_type
-        self.model = SteeredQwen3_5TextModel(args)
-        if not args.tie_word_embeddings:
-            self.lm_head = nn.Linear(args.hidden_size, args.vocab_size, bias=False)
-
-    def __call__(
-        self,
-        inputs: mx.array,
-        cache: Optional[Any] = None,
-        input_embeddings: Optional[mx.array] = None,
-    ) -> mx.array:
-        out = self.model(inputs, cache, input_embeddings=input_embeddings)
-        if self.args.tie_word_embeddings:
-            out = self.model.embed_tokens.as_linear(out)
-        else:
-            out = self.lm_head(out)
-        return out
-
-    @property
-    def layers(self):
-        return self.model.layers
-
-    @property
-    def head_dim(self):
-        return self.args.head_dim
-
-    def make_cache(self):
-        return [ArraysCache(size=2) if l.is_linear else KVCache() for l in self.layers]
-
-    def sanitize(self, weights):
-        # TextModel just passes through — sanitization done at Model level
-        return weights
-
-
-# =====================================================================
-# Public API (required by mlx_lm)
-# =====================================================================
-
-@dataclass
-class ModelArgs(BaseModelArgs):
-    model_type: str = ""
-    text_config: dict = field(default_factory=dict)
-
-    @classmethod
-    def from_dict(cls, params):
-        if "text_config" not in params:
-            return cls(model_type=params["model_type"], text_config=params)
-        return super().from_dict(params)
-
-
-class Model(nn.Module):
-    def __init__(self, config):
-        super().__init__()
-        if isinstance(config, ModelArgs):
-            text_config = config.text_config
-        else:
-            text_config = config.get("text_config", config)
-
-        self.language_model = TextModel(TextModelArgs.from_dict(text_config))
-        self.config = text_config
-
-    def __call__(
-        self,
-        inputs: mx.array,
-        cache: Optional[Any] = None,
-        input_embeddings: Optional[mx.array] = None,
-        **kwargs,
-    ) -> mx.array:
-        return self.language_model(inputs, cache, input_embeddings=input_embeddings)
-
-    @property
-    def layers(self):
-        return self.language_model.model.layers
-
-    @property
-    def head_dim(self):
-        return self.language_model.head_dim
-
-    @property
-    def n_kv_heads(self):
-        return self.language_model.args.num_key_value_heads
-
-    def make_cache(self):
-        return self.language_model.make_cache()
-
-    def sanitize(self, weights):
-        weights = tree_unflatten(list(weights.items()))
-        weights = dict(tree_flatten(weights))
-
-        new_weights = {}
-        for key, value in weights.items():
-            # Handle steering direction
-            if "steering_direction" in key:
-                new_weights[key] = value
-                continue
-            if key.startswith("model.visual"):
-                continue
-            if key.startswith("model.language_model"):
-                key = key.replace("model.language_model", "language_model.model")
-            elif key.startswith("language_model."):
-                pass
-            else:
-                key = "language_model." + key
-            new_weights[key] = value
-
-        for l in range(self.language_model.args.num_hidden_layers):
-            prefix = f"language_model.model.layers.{l}.mlp"
-            gate_up_key = f"{prefix}.experts.gate_up_proj"
-            if gate_up_key in new_weights:
-                gate_up = new_weights.pop(gate_up_key)
-                mid = gate_up.shape[-2] // 2
-                new_weights[f"{prefix}.switch_mlp.gate_proj.weight"] = gate_up[..., :mid, :]
-                new_weights[f"{prefix}.switch_mlp.up_proj.weight"] = gate_up[..., mid:, :]
-                new_weights[f"{prefix}.switch_mlp.down_proj.weight"] = new_weights.pop(
-                    f"{prefix}.experts.down_proj"
-                )
-
-        return new_weights

model.safetensors-00001-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:4fd9489caf88c267ab766c4b055f4fbccdd7bc9a02e437cd5b9c502abaafac8d
-size 1873281886

model.safetensors-00002-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:e70782263c4cbfb6e1e53d77bf573e7a985744daaab60e1c9dc43dd020a3bbc2
-size 1873281886

model.safetensors-00003-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6b88c2bbc3548ac3a1351519e55507870e8acce4f72267e118a5c2d1ad523117
-size 1873281886

model.safetensors-00004-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:1f3410d449acd3b1f482f96400ef741d6e663b3a08dc8fa19356afa41f61116e
-size 1873281880

model.safetensors-00005-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:1dd6176990d66c6a6c3c18f0b8d29664f6123b26001022325c540963fe91ba18
-size 1873281880

model.safetensors-00006-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:d6ccc0a038a141a87d4684b69b94a01be648ab44f9c8b9640d40f5e2583d1f10
-size 1873281886

model.safetensors-00007-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:cc5c7a1d60da308629de9b53171dd82041576ee1c853b4dfe13abf03a55f91a2
-size 1873281886

model.safetensors-00008-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:53f3f3d92b3973c32fe9c77d7cecef88e46bc6d9227df4ee8c6620e0d05fba6e
-size 1873281886

model.safetensors-00009-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:87001a73bf469a0c2ee22d0a82070529cb88761b6f21f83097d91b07b36e48b8
-size 1873281888

model.safetensors-00010-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a712c5c1aec49c4ed6eb14f1b2a638b67ef2b436407adbd5a2e5c07923df87ed
-size 1873281890

model.safetensors-00011-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6180741be19531e04322b9fe2df90f4a3d626de968a9503c3522e9e9bb722134
-size 1873281886

model.safetensors-00012-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:3f7f060173f4365a5abc5e1938c994ffe6e178c8535b17cd0045f92c9a1e4949
-size 1873281886

model.safetensors-00013-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:8e3706807663f41abddb5d69d71039ec29462a164c19891b968336e180e9e2ee
-size 1873281886

model.safetensors-00014-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6065007821c89e6aa632ca6c1b4bdd3066bdf35f82def5ca041f3a56bf9b3563
-size 1873281888

model.safetensors-00015-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:44d273f642424deeac5439c2d9667f0fde423d5c209880b9fde7377e55272790
-size 1873281886

model.safetensors-00016-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:4ea31d6b82d66a98240ca3993d942e1254dfaecf4870a28e4a9ced1886f769a0
-size 1873281880

model.safetensors-00017-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6b2565b899bb52946a4f27d398f77baf0e5e615be3beaf5e0853cd47ddccb7b0
-size 1873281888

model.safetensors-00018-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:c1de4f1bd40d44a17cd4333c3e9af5341be93453d7079157ddf3acba32037043
-size 1873281880

model.safetensors-00019-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:2e2cf6eed761c74dd57ee7e22ed1cf0116439b4ba38b4392b7a52887aba5cc60
-size 1873281882

model.safetensors-00020-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:012e08a7cde12ffeb64fcc847a0a7bb3f2cc1a5bfb1efe34e11f7070a1e01cc2
-size 1873281886

model.safetensors-00021-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:4193ccfcf4ce7598de33fa4086c82106016ff596075dc7a483249747db4a665d
-size 1873281886

model.safetensors-00022-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6fb364d58f5982be87b927474893dce4d9945e3ec10bf6d23c33abe3a747c923
-size 1873281890

model.safetensors-00023-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:34f6bed657bbd8afd462e642061abd16c4387c0391c995ebcc9c359b8c1213be
-size 1873281886

model.safetensors-00024-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:938e72860c920fe7b0e2863c767b34379e3ef7c3ce0b537b231d31e801d2f5ac
-size 1873281886

model.safetensors-00025-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:845a2c39c5ebfd6e6675c719f593aa386502d243d014dc5375d4c5490968f8a9
-size 1873281886

model.safetensors-00026-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:e5b214e6995508a6848b0b7cabe9b34acbe0778e7b39d70107505de444861023
-size 1873281886

model.safetensors-00027-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:07691a0946c0127bedfd0bf291c4d77351e5b177d00f279db47d1bfbf90f970b
-size 1873281886

model.safetensors-00028-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:ea012d4f1dcd54ccc890827e6c56fa7cebc1418f5f6629307dcc87c74366c57a
-size 1873281880

model.safetensors-00029-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f6f7a8327e4ceebd546a43e80c07fad014107fbbc9509c904b85a5400e43d815
-size 1873281880

model.safetensors-00030-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:3529b734384af722552340df5d864a72a0a1b3ef5a3311704050da0eed6aaecb
-size 1873281886

model.safetensors-00031-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:856cc39b31617f34cf4c6b34658638e995ab8169377b911c57840c9f6603f00c
-size 1873281880

model.safetensors-00032-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:67f51aa413fd144cd87e7b0b62c3e2b7ddde987f6658aacbf73f87777fceea07
-size 1873281886

model.safetensors-00033-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:409e04ca2c0d4841347168d86fc4ff6d552cfa8de3f33eb52a8ab9608608c8f3
-size 1873281886

model.safetensors-00034-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:39646063b59b8a7d33b58771468f01449990bd84266111c829f9e24c19cbc814
-size 1873281890

model.safetensors-00035-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:81d5d197efb505a786c8033dcdc51a4df903ce1000da9df49fca75a4ad9a5b22
-size 1873281880

model.safetensors-00036-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:7da8b293e041bed07abf82f40d4e3ca3c0537f2569a540c8e6630be3097ab65f
-size 1873281886

model.safetensors-00037-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:eb5cb8c8d7b4d56b04fa5b38c7b31da7e7f1a283b003ce2d6f6ec1c2062e41bd
-size 1873281890

model.safetensors-00038-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:879bfd1f58659d9a8786f59bbb25800f7d51e0ebf1676905acd6b5390b42880b
-size 1873281886

model.safetensors-00039-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:0a0944911e92352bf5cd50b5726dce9f25306216267fd87f7b630286dc75a314
-size 1873281886

model.safetensors-00040-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:fbaae2eb04be04de9a27505d6d2a72516c1cfab808b9c4355277bc8baa7fa055
-size 1873281888

model.safetensors-00041-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:330e34d7097a2b871e2a482482e49da929c36ea96e15a2b4b768c3958f87d63e
-size 1873281886

model.safetensors-00042-of-00094.safetensors

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:73e4f5e5bc6d10c44e0058368553f8096c63073d8ef671f0f1621d71668c0b6a
-size 1873281886