Instructions to use AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "AutomatosX/AX-gpt-oss-20b-MLX-AXQ-4bit" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

AX-gpt-oss-20b-MLX-AXQ-4bit / axquant_plan.json

AutomatosX

Rebuild from openai/gpt-oss-20b native MXFP4 mixed source (AXQ-4bit); replace mlx-community Q4 re-pack lineage.

b153d25 verified 6 days ago

Raw

History Blame Contribute Delete

369 kB

	{
	"analysis_sha256": "c263d20cb4e55f11624c0ffd8f2b61210e3a86bd761284c50558c6367e710d12",
	"architecture_profile": {
	"adapter_id": "gpt-oss-v1",
	"audio_present": false,
	"config_model_type": "gpt_oss",
	"dense": false,
	"mtp_declared": false,
	"notes": [
	"OpenAI GPT-OSS MoE (model_type=gpt_oss): fused experts via MLX-LM SwitchGLU.",
	"Native MXFP4 expert packs convert with --allow-quantized (dequant → affine).",
	"No declared MTP; development evidence until a revision is certified.",
	"MoE experts quantize as fused switch modules with uniform per-group precision; artifacts are development evidence until certified."
	],
	"optimization_scope": "text-path",
	"product_family": "gpt-oss",
	"support_level": "supported",
	"support_tier": "convertible",
	"text_layer_count": 24,
	"vision_present": false
	},
	"assignments": [
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.0.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.0.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.0.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.0.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.0.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.0.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.0.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.0.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.0.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.1.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.1.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.1.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.1.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.1.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.1.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.1.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.1.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.1.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.10.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.10.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.10.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.10.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.10.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.10.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.10.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.10.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.10.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.11.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.11.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.11.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.11.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.11.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.11.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.11.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.11.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.11.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.12.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.12.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.12.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.12.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.12.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.12.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.12.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.12.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.12.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.13.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.13.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.13.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.13.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.13.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.13.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.13.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.13.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.13.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.14.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.14.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.14.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.14.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.14.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.14.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.14.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.14.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.14.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.15.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.15.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.15.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.15.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.15.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.15.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.15.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.15.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.15.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.16.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.16.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.16.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.16.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.16.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.16.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.16.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.16.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.16.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.17.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.17.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.17.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.17.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.17.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.17.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.17.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.17.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.17.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.self_attn.k_proj.bias"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.18.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.self_attn.q_proj.bias"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.self_attn.v_proj.bias"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.18.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.18.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.18.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.post_attention_layernorm.weight"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.18.self_attn.k_proj.weight"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.18.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.18.self_attn.sinks"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.18.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.18.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.19.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.19.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.19.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.19.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.19.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.19.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.19.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.19.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.19.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.2.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.2.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.2.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.2.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.2.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.2.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.2.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.2.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.2.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.20.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.20.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.20.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.20.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.20.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.20.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.20.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.20.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.20.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.21.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.21.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.21.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.21.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.21.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.21.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.21.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.21.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.21.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.22.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.22.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.22.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.22.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.22.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.22.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.22.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.22.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.22.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.23.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.23.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.23.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.23.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.23.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.23.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.23.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.23.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.23.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.3.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.3.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.3.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.3.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.3.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.3.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.3.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.3.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.3.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.4.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.4.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.4.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.4.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.4.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.4.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.4.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.4.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.4.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.5.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.5.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.5.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.5.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.5.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.5.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.5.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.5.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.5.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.6.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.6.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.6.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.6.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.6.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "lm_head",
	"outlier_strategy": "none",
	"parameters": 579133440,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-lm-head: Protection floor — LM head stays BF16",
	"role": "lm_head",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "lm_head.weight"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.embed_tokens",
	"outlier_strategy": "none",
	"parameters": 579133440,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-embeddings: Embedding floor 8-bit",
	"role": "embedding",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.embed_tokens.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.6.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.6.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.6.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.6.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.7.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.7.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.7.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.7.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.7.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.7.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.7.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.7.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.7.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.8.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.8.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.8.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.8.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.8.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.8.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.8.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.8.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.8.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.input_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.input_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.mlp.experts.down_proj_bias",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.mlp.experts.down_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.mlp.experts.down_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 265420800,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.9.mlp.experts.down_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.mlp.experts.gate_up_proj_bias",
	"outlier_strategy": "none",
	"parameters": 184320,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "expert",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.mlp.experts.gate_up_proj_bias"
	},
	{
	"bits": 4,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.mlp.experts.gate_up_proj_blocks",
	"outlier_strategy": "none",
	"parameters": 530841600,
	"predicted_loss": 0.0,
	"reason": "manual rule experts-4bit: MoE experts/MLP at 4-bit product trunk",
	"role": "expert",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 4.5
	},
	"tensor": "model.layers.9.mlp.experts.gate_up_proj_blocks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.mlp.router.bias",
	"outlier_strategy": "none",
	"parameters": 32,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "router",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.mlp.router.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.mlp.router",
	"outlier_strategy": "none",
	"parameters": 92160,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-routers: Router floor 8-bit for routing stability",
	"role": "router",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.9.mlp.router.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.post_attention_layernorm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.post_attention_layernorm.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.self_attn.k_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.self_attn.k_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.self_attn.k_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.9.self_attn.k_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.self_attn.o_proj.bias",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.self_attn.o_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.self_attn.o_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.9.self_attn.o_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.self_attn.q_proj.bias",
	"outlier_strategy": "none",
	"parameters": 4096,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.self_attn.q_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.self_attn.q_proj",
	"outlier_strategy": "none",
	"parameters": 11796480,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.9.self_attn.q_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.self_attn.sinks",
	"outlier_strategy": "none",
	"parameters": 64,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.self_attn.sinks"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.self_attn.v_proj.bias",
	"outlier_strategy": "none",
	"parameters": 512,
	"predicted_loss": 0.0,
	"reason": "non-quantizable tensor preserved",
	"role": "attention",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.layers.9.self_attn.v_proj.bias"
	},
	{
	"bits": 8,
	"group_size": 64,
	"method": "affine",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.layers.9.self_attn.v_proj",
	"outlier_strategy": "none",
	"parameters": 1474560,
	"predicted_loss": 0.0,
	"reason": "manual rule attention-8bit: Raise attention for general + coding fidelity",
	"role": "attention",
	"scale_strategy": "group-affine",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 8.5
	},
	"tensor": "model.layers.9.self_attn.v_proj.weight"
	},
	{
	"bits": 16,
	"group_size": null,
	"method": "bf16",
	"metrics": {
	"cosine_distance": 0.0,
	"decode_latency_cost": 0.0,
	"hidden_state_error": 0.0,
	"long_context_loss": 0.0,
	"mtp_acceptance_loss": 0.0,
	"output_kl": 0.0,
	"peak_memory_cost": 0.0,
	"prefill_latency_cost": 0.0,
	"task_loss_delta": 0.0,
	"token_disagreement": 0.0
	},
	"module_path": "model.norm",
	"outlier_strategy": "none",
	"parameters": 2880,
	"predicted_loss": 0.0,
	"reason": "manual rule protect-norms: Protection floor — norms stay BF16",
	"role": "norm",
	"scale_strategy": "none",
	"strategy_metadata": {
	"selected_from_candidates": 1,
	"storage_bpw": 16.0
	},
	"tensor": "model.norm.weight"
	}
	],
	"calibration": null,
	"candidate_bits": [
	4,
	8,
	16
	],
	"candidate_group_sizes": [],
	"constraints": {
	"effective_bpw_limit": 5.2,
	"lm_head_min_bits": 16,
	"max_model_size_ratio_to_uniform4": 1.1,
	"minimum_mtp_acceptance_retention": 0.95,
	"minimum_mtp_speedup": 1.2,
	"minimum_quality_retention": 0.98
	},
	"cost_model": "abstract-bpw",
	"created_at": "2026-08-11T22:44:24.531951Z",
	"effective_bpw": 5.055283578663038,
	"evidence_kind": "architecture_prior",
	"global_validation_required": true,
	"group_size": 64,
	"hardware": {
	"name": "ax-engine-apple-silicon-affine-dwq-v3",
	"runtime": "ax-engine",
	"supported_bits": [
	2,
	3,
	4,
	6,
	8,
	16
	],
	"supported_group_sizes": [
	32,
	64,
	128
	],
	"supported_methods": [
	"affine",
	"awq",
	"dwq",
	"gptq",
	"gptq-act",
	"bf16"
	]
	},
	"kernel_latency_host_id": null,
	"kernel_latency_sha256": null,
	"kv_cache": null,
	"method_near_ties": [],
	"method_near_ties_omitted": 0,
	"mtp": {
	"candidate_bits": [
	8,
	16
	],
	"min_bits": 8,
	"mode": "protected",
	"optimize_for_acceptance": true,
	"preserve_external_sidecar": true,
	"protect_norms": true,
	"protect_output_head": true
	},
	"mtp_distribution": {},
	"nominal_bpw": 4.5692953211576715,
	"objective": {
	"cosine_distance": 0.03,
	"decode_latency_cost": 0.07,
	"hidden_state_error": 0.1,
	"long_context_loss": 0.05,
	"mtp_acceptance_loss": 0.22,
	"output_kl": 0.15,
	"peak_memory_cost": 0.03,
	"prefill_latency_cost": 0.03,
	"task_loss_delta": 0.2,
	"token_disagreement": 0.12
	},
	"primary_runtime": "ax-engine",
	"profile": "agent-coding",
	"quantizer": "axquant",
	"random_seed": 20260728,
	"schema_version": "axquant.plan.v1",
	"software_versions": {
	"ax_engine": "6.11.1",
	"axquant": "1.6.2",
	"mlx": "0.32.0",
	"mlx_lm": "0.31.3",
	"pydantic": "2.13.4",
	"python": "3.12.13",
	"safetensors": "0.8.0"
	},
	"source_model": {
	"architecture": "GptOssForCausalLM",
	"format": "mlx",
	"local_path": null,
	"model_id": "openai/gpt-oss-20b",
	"revision": "6cee5e81ee83917806bbde320786a8fb61efebee"
	},
	"status": "planned",
	"target_bpw": 5.2,
	"target_class": "4bit",
	"target_mode": "balanced",
	"warnings": [
	"Manual assignments are unmeasured development evidence.",
	"Conversion requires --allow-unmeasured and cannot pass publication gates."
	],
	"weight_distribution": {
	"4bit": {
	"fraction": 0.9137231396891172,
	"parameters": 19110297600
	},
	"8bit": {
	"fraction": 0.05825337532161521,
	"parameters": 1218355200
	},
	"bf16": {
	"fraction": 0.028023484989267565,
	"parameters": 586104384
	}
	}
	}