[ { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-MQ-IQ4_XS_1-Generic.gguf", "DisplayName": "MQ-IQ4_XS_1", "ProviderSource": "MagicQuant recipe clone", "BaselineFamily": "IQ4_XS", "OriginalReferenceBaseline": "IQ4_NL", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "MQ-IQ4_XS_1", "SourceRun": "Unsloth Dynamic V2", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Generic", "sha256": "123a92c3ba8cd31ed2887bd348682be5b68c977b12cc30e11c632e7ddf899eaa" }, "TensorGroups": { "attention_kv": { "MXFP4": 32 }, "attention_or_ssm_output": { "MXFP4": 64 }, "attention_q_or_qkv": { "MXFP4": 64 }, "embeddings": { "Q4_K": 1 }, "ffn_down": { "MXFP4": 64 }, "ffn_up_gate": { "MXFP4": 128 }, "lm_head": { "Q6_K": 1 }, "mtp_block_64": { "F32": 7, "Q4_K": 6, "Q5_K": 1, "Q8_0": 1 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "MXFP4": 144 } }, "EffectiveTypeCounts": { "F32": 360, "MXFP4": 496, "Q4_K": 7, "Q5_K": 1, "Q6_K": 1, "Q8_0": 1 }, "ChangedTensorCount": 10, "PreservedTensorCount": 856, "NativeMxfp4PreservedCount": 496, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "Q4_K": 6, "Q5_K": 1, "Q8_0": 1 } }, "ActualSizeBytes": 14981769120, "ActualSizeGB": 14.98176912, "ActualSizeGiB": 13.952859789133072, "KldVsNativeMxfp4": 0.00094, "Sha256": "215dc51d41a212d58168ed80e77d0e14ca649d627563c389786263d77d1d4a3d", "VerificationPassed": true, "PublicationStatus": "published" }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-UD-Q4_K_S-Unsloth.gguf", "DisplayName": "UD-Q4_K_S", "ProviderSource": "Unsloth recipe clone", "BaselineFamily": "UD-Unsloth-UD-Q4_K_S", "OriginalReferenceBaseline": "Q4_K_S", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "UD-Unsloth-UD-Q4_K_S", "SourceRun": "Unsloth Dynamic V3", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Unsloth", "sha256": "0ee5b10bd0c2fa2127c6f4b43dbfe1efd71e383b63217af9dade1de36599f1c1" }, "TensorGroups": { "attention_kv": { "MXFP4": 32 }, "attention_or_ssm_output": { "MXFP4": 64 }, "attention_q_or_qkv": { "MXFP4": 64 }, "embeddings": { "Q3_K": 1 }, "ffn_down": { "IQ3_S": 6, "IQ3_XXS": 1, "MXFP4": 56, "Q3_K": 1 }, "ffn_up_gate": { "IQ2_S": 1, "IQ2_XS": 1, "IQ3_S": 9, "IQ3_XXS": 4, "MXFP4": 102, "Q3_K": 11 }, "lm_head": { "Q6_K": 1 }, "mtp_block_64": { "F32": 7, "Q6_K": 6, "Q8_0": 2 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "MXFP4": 144 } }, "EffectiveTypeCounts": { "F32": 360, "IQ2_S": 1, "IQ2_XS": 1, "IQ3_S": 15, "IQ3_XXS": 5, "MXFP4": 462, "Q3_K": 13, "Q6_K": 7, "Q8_0": 2 }, "ChangedTensorCount": 44, "PreservedTensorCount": 822, "NativeMxfp4PreservedCount": 462, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "Q6_K": 6, "Q8_0": 2 } }, "ActualSizeBytes": 14547122080, "ActualSizeGB": 14.54712208, "ActualSizeGiB": 13.548063188791275, "KldVsNativeMxfp4": 0.00345, "Sha256": "b077ade0bd8aab9a1e5386ed09d72b6a24bca62fd71ec16c049b29a740f81ff3", "VerificationPassed": true, "PublicationStatus": "published" }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-UD-IQ4_XS-Unsloth.gguf", "DisplayName": "UD-IQ4_XS", "ProviderSource": "Unsloth recipe clone", "BaselineFamily": "UD-Unsloth-UD-IQ4_XS", "OriginalReferenceBaseline": "IQ4_XS", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "UD-Unsloth-UD-IQ4_XS", "SourceRun": "Unsloth Dynamic V3", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Unsloth", "sha256": "0ee5b10bd0c2fa2127c6f4b43dbfe1efd71e383b63217af9dade1de36599f1c1" }, "TensorGroups": { "attention_kv": { "MXFP4": 32 }, "attention_or_ssm_output": { "MXFP4": 63, "Q3_K": 1 }, "attention_q_or_qkv": { "IQ3_S": 8, "IQ3_XXS": 3, "MXFP4": 52, "Q2_K": 1 }, "embeddings": { "Q3_K": 1 }, "ffn_down": { "IQ2_S": 1, "IQ3_S": 9, "IQ3_XXS": 5, "MXFP4": 48, "Q3_K": 1 }, "ffn_up_gate": { "IQ2_S": 6, "IQ2_XS": 2, "IQ3_S": 28, "IQ3_XXS": 9, "MXFP4": 78, "Q3_K": 5 }, "lm_head": { "Q5_K": 1 }, "mtp_block_64": { "F32": 7, "Q6_K": 6, "Q8_0": 2 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "IQ3_S": 1, "MXFP4": 142, "Q3_K": 1 } }, "EffectiveTypeCounts": { "F32": 360, "IQ2_S": 7, "IQ2_XS": 2, "IQ3_S": 46, "IQ3_XXS": 17, "MXFP4": 415, "Q2_K": 1, "Q3_K": 9, "Q5_K": 1, "Q6_K": 6, "Q8_0": 2 }, "ChangedTensorCount": 91, "PreservedTensorCount": 775, "NativeMxfp4PreservedCount": 415, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "Q6_K": 6, "Q8_0": 2 } }, "ActualSizeBytes": 13887481760, "ActualSizeGB": 13.88748176, "ActualSizeGiB": 12.933725267648697, "KldVsNativeMxfp4": 0.009018, "Sha256": "892cd2ac8aebc6f4d7cdf4d5fb0b417507a8ba9df7f038d45b9185835122cb03", "VerificationPassed": true, "PublicationStatus": "published" }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-UD-Q3_K_XL-Unsloth.gguf", "DisplayName": "UD-Q3_K_XL", "ProviderSource": "Unsloth recipe clone", "BaselineFamily": "UD-Unsloth-UD-Q3_K_XL", "OriginalReferenceBaseline": "IQ3_M", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "UD-Unsloth-UD-Q3_K_XL", "SourceRun": "Unsloth Dynamic V3", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Unsloth", "sha256": "0ee5b10bd0c2fa2127c6f4b43dbfe1efd71e383b63217af9dade1de36599f1c1" }, "TensorGroups": { "attention_kv": { "MXFP4": 32 }, "attention_or_ssm_output": { "IQ3_S": 4, "IQ3_XXS": 2, "MXFP4": 54, "Q3_K": 4 }, "attention_q_or_qkv": { "IQ3_S": 29, "IQ3_XXS": 8, "MXFP4": 24, "Q2_K": 3 }, "embeddings": { "Q3_K": 1 }, "ffn_down": { "IQ2_S": 4, "IQ2_XS": 1, "IQ3_S": 17, "IQ3_XXS": 4, "MXFP4": 35, "Q3_K": 3 }, "ffn_up_gate": { "IQ2_S": 10, "IQ2_XS": 3, "IQ2_XXS": 2, "IQ3_S": 53, "IQ3_XXS": 15, "MXFP4": 41, "Q3_K": 4 }, "lm_head": { "Q5_K": 1 }, "mtp_block_64": { "F32": 7, "Q6_K": 6, "Q8_0": 2 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "IQ2_S": 1, "IQ3_S": 8, "IQ3_XXS": 5, "MXFP4": 130 } }, "EffectiveTypeCounts": { "F32": 360, "IQ2_S": 15, "IQ2_XS": 4, "IQ2_XXS": 2, "IQ3_S": 111, "IQ3_XXS": 34, "MXFP4": 316, "Q2_K": 3, "Q3_K": 12, "Q5_K": 1, "Q6_K": 6, "Q8_0": 2 }, "ChangedTensorCount": 190, "PreservedTensorCount": 676, "NativeMxfp4PreservedCount": 316, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "Q6_K": 6, "Q8_0": 2 } }, "ActualSizeBytes": 13026502560, "ActualSizeGB": 13.02650256, "ActualSizeGiB": 12.131875902414322, "KldVsNativeMxfp4": 0.022344, "Sha256": "149e7378c85768623e649afea5c9184ed1be05532948f841a39edf4870b88e63", "VerificationPassed": true, "PublicationStatus": "published" }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-UD-IQ3_S-Unsloth.gguf", "DisplayName": "UD-IQ3_S", "ProviderSource": "Unsloth recipe clone", "BaselineFamily": "UD-Unsloth-UD-IQ3_S", "OriginalReferenceBaseline": "IQ3_S", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "UD-Unsloth-UD-IQ3_S", "SourceRun": "Unsloth Dynamic V3", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Unsloth", "sha256": "0ee5b10bd0c2fa2127c6f4b43dbfe1efd71e383b63217af9dade1de36599f1c1" }, "TensorGroups": { "attention_kv": { "MXFP4": 32 }, "attention_or_ssm_output": { "IQ3_S": 12, "IQ3_XXS": 3, "MXFP4": 43, "Q3_K": 6 }, "attention_q_or_qkv": { "IQ2_XXS": 1, "IQ3_S": 32, "IQ3_XXS": 21, "MXFP4": 4, "Q2_K": 6 }, "embeddings": { "Q3_K": 1 }, "ffn_down": { "IQ2_S": 6, "IQ2_XS": 1, "IQ2_XXS": 2, "IQ3_S": 23, "IQ3_XXS": 10, "MXFP4": 17, "Q3_K": 5 }, "ffn_up_gate": { "IQ1_S": 2, "IQ2_S": 14, "IQ2_XS": 10, "IQ2_XXS": 9, "IQ3_S": 45, "IQ3_XXS": 24, "MXFP4": 21, "Q3_K": 3 }, "lm_head": { "Q5_K": 1 }, "mtp_block_64": { "F32": 7, "Q6_K": 6, "Q8_0": 2 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "IQ2_S": 1, "IQ2_XS": 1, "IQ3_S": 15, "IQ3_XXS": 19, "MXFP4": 107, "Q3_K": 1 } }, "EffectiveTypeCounts": { "F32": 360, "IQ1_S": 2, "IQ2_S": 21, "IQ2_XS": 12, "IQ2_XXS": 12, "IQ3_S": 127, "IQ3_XXS": 77, "MXFP4": 224, "Q2_K": 6, "Q3_K": 16, "Q5_K": 1, "Q6_K": 6, "Q8_0": 2 }, "ChangedTensorCount": 282, "PreservedTensorCount": 584, "NativeMxfp4PreservedCount": 224, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "Q6_K": 6, "Q8_0": 2 } }, "ActualSizeBytes": 11988330400, "ActualSizeGB": 11.9883304, "ActualSizeGiB": 11.16500273346901, "KldVsNativeMxfp4": 0.042218, "Sha256": "e165d77ffd43985ebee7763e6a9642cdc8e6168ffed5017d954f6c2fc059f719", "VerificationPassed": true, "PublicationStatus": "published" }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-MQ-IQ2_M_1-Generic.gguf", "DisplayName": "MQ-IQ2_M_1", "ProviderSource": "MagicQuant recipe clone", "BaselineFamily": "IQ2_M", "OriginalReferenceBaseline": "IQ3_XXS", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "MQ-IQ2_M_1", "SourceRun": "Unsloth Dynamic V2", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Generic", "sha256": "123a92c3ba8cd31ed2887bd348682be5b68c977b12cc30e11c632e7ddf899eaa" }, "TensorGroups": { "attention_kv": { "MXFP4": 32 }, "attention_or_ssm_output": { "IQ3_S": 16, "MXFP4": 48 }, "attention_q_or_qkv": { "IQ3_S": 48, "MXFP4": 16 }, "embeddings": { "Q2_K": 1 }, "ffn_down": { "IQ3_S": 64 }, "ffn_up_gate": { "IQ3_S": 64, "IQ3_XXS": 64 }, "lm_head": { "Q5_K": 1 }, "mtp_block_64": { "F32": 7, "IQ3_S": 3, "IQ4_XS": 3, "Q4_K": 2 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "IQ1_M": 48, "IQ3_XXS": 48, "MXFP4": 48 } }, "EffectiveTypeCounts": { "F32": 360, "IQ1_M": 48, "IQ3_S": 195, "IQ3_XXS": 112, "IQ4_XS": 3, "MXFP4": 144, "Q2_K": 1, "Q4_K": 2, "Q5_K": 1 }, "ChangedTensorCount": 362, "PreservedTensorCount": 504, "NativeMxfp4PreservedCount": 144, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "IQ3_S": 3, "IQ4_XS": 3, "Q4_K": 2 } }, "ActualSizeBytes": 11913885600, "ActualSizeGB": 11.9138856, "ActualSizeGiB": 11.095670610666275, "KldVsNativeMxfp4": 0.058879, "Sha256": "a242cddbdf9baaad660e96d26be99245f2410c154c3a0fa0c6451e3b7109eb01", "VerificationPassed": true, "PublicationStatus": "published" }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-UD-IQ3_XXS-Unsloth.gguf", "DisplayName": "UD-IQ3_XXS", "ProviderSource": "Unsloth recipe clone", "BaselineFamily": "UD-Unsloth-UD-IQ3_XXS", "OriginalReferenceBaseline": "IQ3_XXS", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "UD-Unsloth-UD-IQ3_XXS", "SourceRun": "Unsloth Dynamic V3", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Unsloth", "sha256": "0ee5b10bd0c2fa2127c6f4b43dbfe1efd71e383b63217af9dade1de36599f1c1" }, "TensorGroups": { "attention_kv": { "IQ3_S": 5, "MXFP4": 27 }, "attention_or_ssm_output": { "IQ2_S": 1, "IQ2_XXS": 1, "IQ3_S": 26, "IQ3_XXS": 10, "MXFP4": 23, "Q3_K": 3 }, "attention_q_or_qkv": { "IQ2_S": 3, "IQ2_XXS": 5, "IQ3_S": 21, "IQ3_XXS": 27, "MXFP4": 2, "Q2_K": 6 }, "embeddings": { "Q2_K": 1 }, "ffn_down": { "IQ1_M": 2, "IQ1_S": 1, "IQ2_S": 7, "IQ2_XS": 4, "IQ2_XXS": 4, "IQ3_S": 20, "IQ3_XXS": 15, "MXFP4": 9, "Q2_K": 1, "Q3_K": 1 }, "ffn_up_gate": { "IQ1_M": 1, "IQ1_S": 7, "IQ2_S": 18, "IQ2_XS": 7, "IQ2_XXS": 13, "IQ3_S": 21, "IQ3_XXS": 44, "MXFP4": 15, "Q3_K": 2 }, "lm_head": { "Q4_K": 1 }, "mtp_block_64": { "F32": 7, "Q6_K": 6, "Q8_0": 2 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "IQ2_S": 6, "IQ2_XS": 1, "IQ2_XXS": 1, "IQ3_S": 13, "IQ3_XXS": 24, "MXFP4": 97, "Q2_K": 2 } }, "EffectiveTypeCounts": { "F32": 360, "IQ1_M": 3, "IQ1_S": 8, "IQ2_S": 35, "IQ2_XS": 12, "IQ2_XXS": 24, "IQ3_S": 106, "IQ3_XXS": 120, "MXFP4": 173, "Q2_K": 10, "Q3_K": 6, "Q4_K": 1, "Q6_K": 6, "Q8_0": 2 }, "ChangedTensorCount": 333, "PreservedTensorCount": 533, "NativeMxfp4PreservedCount": 173, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "Q6_K": 6, "Q8_0": 2 } }, "ActualSizeBytes": 10903975840, "ActualSizeGB": 10.90397584, "ActualSizeGiB": 10.155118852853775, "KldVsNativeMxfp4": 0.072327, "Sha256": "0cf1c2f3106833a487ccd6b1c89b185ad96dc9e9146b7d5367ab9ec797579769", "VerificationPassed": true, "PublicationStatus": "published" }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-MQ-IQ2_M_2-Generic.gguf", "DisplayName": "MQ-IQ2_M_2", "ProviderSource": "MagicQuant recipe clone", "BaselineFamily": "IQ2_M", "OriginalReferenceBaseline": "IQ2_M", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "MQ-IQ2_M_2", "SourceRun": "Unsloth Dynamic V2", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Generic", "sha256": "123a92c3ba8cd31ed2887bd348682be5b68c977b12cc30e11c632e7ddf899eaa" }, "TensorGroups": { "attention_kv": { "MXFP4": 32 }, "attention_or_ssm_output": { "IQ3_S": 64 }, "attention_q_or_qkv": { "IQ3_S": 16, "IQ3_XXS": 48 }, "embeddings": { "Q2_K": 1 }, "ffn_down": { "IQ3_XXS": 64 }, "ffn_up_gate": { "IQ3_XXS": 128 }, "lm_head": { "Q3_K": 1 }, "mtp_block_64": { "F32": 7, "IQ3_S": 2, "IQ4_XS": 4, "Q5_K": 1, "Q6_K": 1 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "IQ1_M": 48, "IQ3_XXS": 48, "MXFP4": 48 } }, "EffectiveTypeCounts": { "F32": 360, "IQ1_M": 48, "IQ3_S": 82, "IQ3_XXS": 288, "IQ4_XS": 4, "MXFP4": 80, "Q2_K": 1, "Q3_K": 1, "Q5_K": 1, "Q6_K": 1 }, "ChangedTensorCount": 426, "PreservedTensorCount": 440, "NativeMxfp4PreservedCount": 80, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "IQ3_S": 2, "IQ4_XS": 4, "Q5_K": 1, "Q6_K": 1 } }, "ActualSizeBytes": 10691495840, "ActualSizeGB": 10.69149584, "ActualSizeGiB": 9.957231432199478, "KldVsNativeMxfp4": 0.102424, "Sha256": "a7ac992cc59c3ec95105d56c8090fd1208f0ee65bc7571d33c4202afaa8c5dc8", "VerificationPassed": true, "PublicationStatus": "published" }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-UD-Q2_K_XL-Unsloth.gguf", "DisplayName": "UD-Q2_K_XL", "ProviderSource": "Unsloth recipe clone", "BaselineFamily": "UD-Unsloth-UD-Q2_K_XL", "OriginalReferenceBaseline": "IQ2_M", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "UD-Unsloth-UD-Q2_K_XL", "SourceRun": "Unsloth Dynamic V3", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Unsloth", "sha256": "0ee5b10bd0c2fa2127c6f4b43dbfe1efd71e383b63217af9dade1de36599f1c1" }, "TensorGroups": { "attention_kv": { "IQ3_S": 7, "IQ3_XXS": 1, "MXFP4": 23, "Q3_K": 1 }, "attention_or_ssm_output": { "IQ2_S": 5, "IQ2_XS": 5, "IQ2_XXS": 1, "IQ3_S": 21, "IQ3_XXS": 21, "MXFP4": 9, "Q3_K": 2 }, "attention_q_or_qkv": { "IQ1_M": 1, "IQ2_S": 12, "IQ2_XS": 6, "IQ2_XXS": 9, "IQ3_S": 4, "IQ3_XXS": 22, "Q2_K": 10 }, "embeddings": { "Q2_K": 1 }, "ffn_down": { "IQ1_S": 3, "IQ2_S": 16, "IQ2_XS": 5, "IQ2_XXS": 8, "IQ3_S": 11, "IQ3_XXS": 14, "MXFP4": 5, "Q2_K": 1, "Q3_K": 1 }, "ffn_up_gate": { "IQ1_S": 17, "IQ2_S": 26, "IQ2_XS": 14, "IQ2_XXS": 24, "IQ3_S": 12, "IQ3_XXS": 29, "MXFP4": 4, "Q2_K": 1, "Q3_K": 1 }, "lm_head": { "Q4_K": 1 }, "mtp_block_64": { "F32": 7, "Q6_K": 6, "Q8_0": 2 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "IQ2_S": 8, "IQ2_XS": 4, "IQ2_XXS": 6, "IQ3_S": 2, "IQ3_XXS": 25, "MXFP4": 96, "Q2_K": 3 } }, "EffectiveTypeCounts": { "F32": 360, "IQ1_M": 1, "IQ1_S": 20, "IQ2_S": 67, "IQ2_XS": 34, "IQ2_XXS": 48, "IQ3_S": 57, "IQ3_XXS": 112, "MXFP4": 137, "Q2_K": 16, "Q3_K": 5, "Q4_K": 1, "Q6_K": 6, "Q8_0": 2 }, "ChangedTensorCount": 369, "PreservedTensorCount": 497, "NativeMxfp4PreservedCount": 137, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "Q6_K": 6, "Q8_0": 2 } }, "ActualSizeBytes": 9809893280, "ActualSizeGB": 9.80989328, "ActualSizeGiB": 9.136175066232681, "KldVsNativeMxfp4": 0.111015, "Sha256": "9b899d16649813748689f2b99407b299cc9eeeae66289cf06066b050f10f3675", "VerificationPassed": true, "PublicationStatus": "published" }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-UD-IQ2_XXS-Unsloth.gguf", "DisplayName": "UD-IQ2_XXS", "ProviderSource": "Unsloth recipe clone", "BaselineFamily": "UD-Unsloth-UD-IQ2_XXS", "OriginalReferenceBaseline": "IQ2_XXS", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "UD-Unsloth-UD-IQ2_XXS", "SourceRun": "Unsloth Dynamic V2", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Unsloth", "sha256": "0ee5b10bd0c2fa2127c6f4b43dbfe1efd71e383b63217af9dade1de36599f1c1" }, "TensorGroups": { "attention_kv": { "IQ3_XXS": 16, "MXFP4": 16 }, "attention_or_ssm_output": { "IQ2_S": 16, "IQ3_XXS": 48 }, "attention_q_or_qkv": { "IQ2_XS": 48, "IQ3_XXS": 16 }, "embeddings": { "Q2_K": 1 }, "ffn_down": { "IQ2_S": 64 }, "ffn_up_gate": { "IQ2_S": 128 }, "lm_head": { "Q3_K": 1 }, "mtp_block_64": { "F32": 7, "IQ4_XS": 7, "Q3_K": 1 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "IQ1_M": 48, "IQ2_XXS": 48, "MXFP4": 48 } }, "EffectiveTypeCounts": { "F32": 360, "IQ1_M": 48, "IQ2_S": 208, "IQ2_XS": 48, "IQ2_XXS": 48, "IQ3_XXS": 80, "IQ4_XS": 7, "MXFP4": 64, "Q2_K": 1, "Q3_K": 2 }, "ChangedTensorCount": 442, "PreservedTensorCount": 424, "NativeMxfp4PreservedCount": 64, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": false, "effectiveTypeCounts": { "F32": 7, "IQ4_XS": 7, "Q3_K": 1 } }, "ActualSizeBytes": 9013733280, "ActualSizeGB": 9.01373328, "ActualSizeGiB": 8.394693285226822, "KldVsNativeMxfp4": 1.172122, "Sha256": "389126864bd6e3fb51639c9727567c91ebcec5e19b5c0ec072ee24d405c97f70", "VerificationPassed": true, "PublicationStatus": "removed", "RemovalReasonCode": "FAILED_QUALITY_FLOOR_AND_STRICT_DOMINANCE", "RemovalReason": "Removed from distribution after verification: KLD 1.172122 and PPL 17.87418 versus native PPL 5.801511. It is also strictly dominated by MQ-IQ2_XXS_1, which is smaller (8.22 GB) and has much lower KLD (0.321797)." }, { "ExportedFileName": "Qwen3.8-27B-Quark-MXFP4-MQ-IQ2_XXS_1-Unsloth.gguf", "DisplayName": "MQ-IQ2_XXS_1", "ProviderSource": "MagicQuant recipe clone", "BaselineFamily": "IQ2_XXS", "OriginalReferenceBaseline": "IQ2_XXS", "SourceRecipeRepository": "magiccodingman/Qwen3.8-27B-MagicQuant-GGUF", "SourceRecipeName": "MQ-IQ2_XXS_1", "SourceRun": "Unsloth Dynamic V3", "Adaptation": { "mode": "strict-downward-only", "isPureClone": false, "bf16Intermediate": false, "rule": "Never increase a source tensor's storage precision merely to match the recipe; preserve source on ties or upward requests." }, "Imatrix": { "label": "Unsloth", "sha256": "0ee5b10bd0c2fa2127c6f4b43dbfe1efd71e383b63217af9dade1de36599f1c1" }, "TensorGroups": { "attention_kv": { "IQ2_XS": 4, "IQ3_S": 6, "IQ3_XXS": 9, "MXFP4": 11, "Q2_K": 2 }, "attention_or_ssm_output": { "IQ1_M": 3, "IQ1_S": 7, "IQ2_S": 12, "IQ2_XS": 13, "IQ2_XXS": 11, "IQ3_XXS": 18 }, "attention_q_or_qkv": { "IQ1_M": 5, "IQ1_S": 5, "IQ2_S": 6, "IQ2_XS": 19, "IQ2_XXS": 20, "IQ3_S": 1, "IQ3_XXS": 5, "Q2_K": 3 }, "embeddings": { "Q2_K": 1 }, "ffn_down": { "IQ1_M": 1, "IQ1_S": 20, "IQ2_S": 7, "IQ2_XS": 2, "IQ2_XXS": 26, "IQ3_S": 3, "IQ3_XXS": 5 }, "ffn_up_gate": { "IQ1_M": 6, "IQ1_S": 53, "IQ2_S": 9, "IQ2_XS": 5, "IQ2_XXS": 45, "IQ3_XXS": 8, "MXFP4": 1, "Q3_K": 1 }, "lm_head": { "Q3_K": 1 }, "mtp_block_64": { "BF16": 8, "F32": 7 }, "norms_biases_and_state": { "F32": 305 }, "other": { "F32": 48, "IQ1_M": 4, "IQ1_S": 6, "IQ2_S": 6, "IQ2_XS": 3, "IQ2_XXS": 29, "MXFP4": 96 } }, "EffectiveTypeCounts": { "BF16": 8, "F32": 360, "IQ1_M": 19, "IQ1_S": 91, "IQ2_S": 40, "IQ2_XS": 46, "IQ2_XXS": 131, "IQ3_S": 10, "IQ3_XXS": 45, "MXFP4": 108, "Q2_K": 6, "Q3_K": 2 }, "ChangedTensorCount": 390, "PreservedTensorCount": 476, "NativeMxfp4PreservedCount": 108, "MtpBlock": { "present": true, "sourceLayerCount": 1, "fullyPreservedAtSourceTypes": true, "effectiveTypeCounts": { "BF16": 8, "F32": 7 } }, "ActualSizeBytes": 8223819680, "ActualSizeGB": 8.22381968, "ActualSizeGiB": 7.659028917551041, "KldVsNativeMxfp4": 0.321797, "Sha256": "84ab6e9cea4e1f7091e00a0b115ffa245e95f8b59e87e6e8ffdfdf3ee71dc441", "VerificationPassed": true, "PublicationStatus": "published" } ]