models : add DeepSeek V4 Flash conversions

models/deepseek-v4-flash-0731/README.md

@@ -0,0 +1,27 @@
+---
+license: mit
+pipeline_tag: text-generation
+tags:
+- gguf
+- quantized
+base_model:
+- deepseek-ai/DeepSeek-V4-Flash-0731
+---
+
+# DeepSeek-V4-Flash-0731
+
+Run with https://llama.app
+
+```bash
+llama serve -hf __owner__/DeepSeek-V4-Flash-0731-GGUF
+```
+
+### Source models
+- https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
+
+### TODOs
+
+- add info
+
+> [!IMPORTANT]
+> This model is automatically converted using https://github.com/ggml-org/convert

models/deepseek-v4-flash-0731/config.sh

@@ -0,0 +1,3 @@
+DISPLAY_NAME="DeepSeek-V4-Flash-0731"
+DEST_REPO="DeepSeek-V4-Flash-0731-GGUF"
+DEP_PRIMARY="deepseek-ai/DeepSeek-V4-Flash-0731"

models/deepseek-v4-flash-0731/convert.sh

@@ -0,0 +1,32 @@
+#!/bin/bash
+set -euox pipefail
+
+OUTPUT_DIR="$1"
+LLAMA_CPP="$2"
+
+DISPLAY_NAME="DeepSeek-V4-Flash-0731"
+QUANTIZE="$LLAMA_CPP/build/bin/llama-quantize"
+
+# --- Conversions ---
+
+# Main model: BF16 (intermediate for quantization only)
+python3 "$LLAMA_CPP/convert_hf_to_gguf.py" "$PATH_PRIMARY" \
+    --outtype bf16 --outfile "$OUTPUT_DIR/${DISPLAY_NAME}-BF16.gguf" --no-mtp --model-name "$DISPLAY_NAME"
+
+# DSpark sidecar: BF16
+python3 "$LLAMA_CPP/convert_hf_to_gguf.py" "$PATH_PRIMARY" \
+    --outtype bf16 --outfile "$OUTPUT_DIR/dspark-${DISPLAY_NAME}-BF16.gguf" --dspark --target-model-dir "$PATH_PRIMARY" --model-name "$DISPLAY_NAME"
+
+# --- Quantizations ---
+
+# Main model: MXFP4_MOE
+"$QUANTIZE" "$OUTPUT_DIR/${DISPLAY_NAME}-BF16.gguf" "$OUTPUT_DIR/${DISPLAY_NAME}-MXFP4.gguf" MXFP4_MOE 1>&2
+
+# DSpark sidecar: MXFP4
+"$QUANTIZE" "$OUTPUT_DIR/dspark-${DISPLAY_NAME}-BF16.gguf" "$OUTPUT_DIR/dspark-${DISPLAY_NAME}-MXFP4.gguf" MXFP4_MOE 1>&2
+
+# --- Produced files ---
+
+echo "${DISPLAY_NAME}-MXFP4.gguf" >> "$OUTPUT_DIR/.produced_files"
+echo "dspark-${DISPLAY_NAME}-BF16.gguf" >> "$OUTPUT_DIR/.produced_files"
+echo "dspark-${DISPLAY_NAME}-MXFP4.gguf" >> "$OUTPUT_DIR/.produced_files"

models/deepseek-v4-flash/README.md

@@ -0,0 +1,27 @@
+---
+license: mit
+pipeline_tag: text-generation
+tags:
+- gguf
+- quantized
+base_model:
+- deepseek-ai/DeepSeek-V4-Flash
+---
+
+# DeepSeek-V4-Flash
+
+Run with https://llama.app
+
+```bash
+llama serve -hf __owner__/DeepSeek-V4-Flash-GGUF
+```
+
+### Source models
+- https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash
+
+### TODOs
+
+- add info
+
+> [!IMPORTANT]
+> This model is automatically converted using https://github.com/ggml-org/convert

models/deepseek-v4-flash/config.sh

@@ -0,0 +1,3 @@
+DISPLAY_NAME="DeepSeek-V4-Flash"
+DEST_REPO="DeepSeek-V4-Flash-GGUF"
+DEP_PRIMARY="deepseek-ai/DeepSeek-V4-Flash"

models/deepseek-v4-flash/convert.sh

@@ -0,0 +1,32 @@
+#!/bin/bash
+set -euox pipefail
+
+OUTPUT_DIR="$1"
+LLAMA_CPP="$2"
+
+DISPLAY_NAME="DeepSeek-V4-Flash"
+QUANTIZE="$LLAMA_CPP/build/bin/llama-quantize"
+
+# --- Conversions ---
+
+# Main model: BF16 (intermediate for quantization only)
+python3 "$LLAMA_CPP/convert_hf_to_gguf.py" "$PATH_PRIMARY" \
+    --outtype bf16 --outfile "$OUTPUT_DIR/${DISPLAY_NAME}-BF16.gguf" --no-mtp --model-name "$DISPLAY_NAME"
+
+# MTP sidecar: BF16
+python3 "$LLAMA_CPP/convert_hf_to_gguf.py" "$PATH_PRIMARY" \
+    --outtype bf16 --outfile "$OUTPUT_DIR/mtp-${DISPLAY_NAME}-BF16.gguf" --mtp --model-name "$DISPLAY_NAME"
+
+# --- Quantizations ---
+
+# Main model: MXFP4_MOE
+"$QUANTIZE" "$OUTPUT_DIR/${DISPLAY_NAME}-BF16.gguf" "$OUTPUT_DIR/${DISPLAY_NAME}-MXFP4.gguf" MXFP4_MOE 1>&2
+
+# MTP sidecar: MXFP4
+"$QUANTIZE" "$OUTPUT_DIR/mtp-${DISPLAY_NAME}-BF16.gguf" "$OUTPUT_DIR/mtp-${DISPLAY_NAME}-MXFP4.gguf" MXFP4_MOE 1>&2
+
+# --- Produced files ---
+
+echo "${DISPLAY_NAME}-MXFP4.gguf" >> "$OUTPUT_DIR/.produced_files"
+echo "mtp-${DISPLAY_NAME}-BF16.gguf" >> "$OUTPUT_DIR/.produced_files"
+echo "mtp-${DISPLAY_NAME}-MXFP4.gguf" >> "$OUTPUT_DIR/.produced_files"