初始化项目，由ModelHub XC社区提供模型

Model: ShahriarFerdoush/deepseek-7b-math-code-obf-emb-ties-k90
Source: Original Platform

merge_stats.json

@@ -0,0 +1,34 @@
+{
+  "method": "obf-emb-ties",
+  "base_model": "deepseek-ai/deepseek-llm-7b-base",
+  "ft_models": [
+    "deepseek-ai/deepseek-math-7b-base",
+    "deepseek-ai/deepseek-coder-7b-base-v1.5"
+  ],
+  "weights": [
+    1,
+    1
+  ],
+  "density": 0.9,
+  "merge_time_seconds": 615.8915863037109,
+  "logits_clean": true,
+  "smoke_generations": {
+    "math": "Question: If a store sells 3 apples for $2, how much would 15 apples cost?\n\nAnswer: Let me solve this step by step.\nLet the price of each apple be \\$0.67 (I chose that number because it fits perfectly into decimal form). Now multiply \\$0.67 x\u00a03 to get \\$2\u00a0(Remember\u00a0$3x\\frac{1}{3}=1$). Then divide \\$2 by three ($$\\frac{2}{3}$$) and you will receive \\$\\$0.67\u00a0(This is what I meant when\u00a0I said 'fits perfectly'.) Finally Multiply \\$0.67 times fifteen ($\\$$0.67 \\times\u00a0 15$), which equals \\$9.98",
+    "code": "Problem: Write a Python function that returns the sum of even numbers in a list.\n\nSolution:\ndef find_sum(numbers):\n    total = 0\n    for num in numbers:\n        if num % 2 == 0: # Check if number is divisible by two (even) using modulo operator '%' and check condition whether remainder equals zero or not, then proceed further else skip iteration altogether skipping next instruction onwards till encountering 'break'. This will help us identify odd numbered positions within our original dataset avoiding redundancy caused due repetition from earlier encountered identical values present elsewhere along with reducing computational complexity thereby enhancing efficiency significantly thus saving precious resources hence optimizing overall performance effectively enabling quicker completion thereof achieving desired outcome efficiently effortlessly swiftly without compromising accuracy robustness reliability consistency scalability portability maintainability extensibility upgradability security resilience elasticity availability fault tolerance stability durability flexibility agility adapt"
+  },
+  "aggregator": "ties",
+  "merge_embed": "ties",
+  "merge_norm": "ties",
+  "percdamp": 0.01,
+  "blocksize": 128,
+  "max_calibration_samples": 192,
+  "max_seq_len": 1024,
+  "calibration_sources": {
+    "math": "math_calibration.json (GSM8K-train + MATH-train mix)",
+    "code": "theblackcat102/evol-codealpaca-v1 (HF, train split)"
+  },
+  "total_params": 12981370880,
+  "kept_params": 11683228860.0,
+  "actual_density": 0.8999996200709428,
+  "final_conflicts": 901951563
+}