[Mini-VLLM] Cut non-OpenAI surfaces, dead kernels, Mistral stack

- Remove Anthropic, Cohere, SageMaker, decisions, structured-decisions,
  rl, render-launcher and batch entrypoints; keep scale-out token routes
- Drop the Swagger static bundle and the /docs, /redoc routes
- Delete device-keyed tuning configs for fused_moe, quant utils, mamba
  select-state and helion; no sm86 entry ever existed for any of them
- Remove ten unused rope variants plus the fbgemm_fp8 and fp_quant
  quantizers and their config, metrics and loader plumbing
- Cut the CPU device build: csrc/cpu, cpu_extension.cmake, cpu
  requirements and the setup.py cpu target path
- Delete 29 environment variables nothing reads and 14 chat-template
  fallbacks for models this fork no longer registers
- Remove nine dead CUDA ops (deepseek_v4, kimi_k3, minimax, muse_spark,
  dsv3 gemm) with their kernels, bindings, wrappers and CMake entries
- Delete the connector test suites that targeted removed connectors
- Cut the Mistral tokenizer and parser stack, its tests, and the
  mistral_common and ijson dependency pins
- Restore the Qwen4Exp registry rows and nvidia/mtp.py that an earlier
  cut deleted while keeping the model tree

Co-authored-by: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>
This commit is contained in:
2026-10-19 17:10:56 +07:00
co-authored by Hinata
parent bd70a61c20
commit 2066d3143f
1086 changed files with 122 additions and 291919 deletions
-177
View File
@@ -172,9 +172,6 @@ torch::stable::Tensor awq_dequantize(torch::stable::Tensor _kernel,
int64_t split_k_iters, int64_t thx,
int64_t thy);
// DSV3 fused A GEMM: conditionally compiled so declaration and impl
// registration are in the source file (dsv3_fused_a_gemm.cu)
#endif
// CPU tensor -> CUDA UVA view (shared CUDA/ROCm)
@@ -256,164 +253,6 @@ void fused_qk_norm_rope(torch::stable::Tensor& qkv, int64_t num_heads_q,
torch::stable::Tensor& position_ids,
int64_t forced_token_heads_per_warp);
void fused_deepseek_v4_kv_rope_insert(
torch::stable::Tensor const& kv, torch::stable::Tensor& k_cache,
torch::stable::Tensor const& slot_mapping,
torch::stable::Tensor const& position_ids,
torch::stable::Tensor const& cos_sin_cache, int64_t cache_block_size,
std::optional<torch::stable::Tensor> fp8_scale, bool kv_mxfp8);
torch::stable::Tensor fused_deepseek_v4_qnorm_rope_kv_rope_quant_insert(
torch::stable::Tensor const& q_in, torch::stable::Tensor const& kv,
torch::stable::Tensor& k_cache, torch::stable::Tensor const& slot_mapping,
torch::stable::Tensor const& position_ids,
torch::stable::Tensor const& cos_sin_cache, int64_t q_head_padded,
double eps, int64_t cache_block_size, bool apply_q_norm, bool kv_mxfp8,
bool apply_q_rope, bool is_q_interleaved);
void fused_deepseek_v4_qnorm_rope_kv_rope_full_cache_bf16_insert(
torch::stable::Tensor& q, torch::stable::Tensor const& kv,
torch::stable::Tensor& k_cache, torch::stable::Tensor const& slot_mapping,
torch::stable::Tensor const& position_ids,
torch::stable::Tensor const& cos_sin_cache, double eps,
int64_t cache_block_size, bool apply_q_norm);
void fused_kimi_k3_mla_key_concat_kv_cache_insert(
torch::stable::Tensor& q, torch::stable::Tensor const& k_nope,
torch::stable::Tensor const& k_pe, torch::stable::Tensor const& kv_c_normed,
torch::stable::Tensor& k_out, torch::stable::Tensor& k_cache,
torch::stable::Tensor const& slot_mapping, int64_t cache_block_size,
std::optional<torch::stable::Tensor> position_ids,
std::optional<torch::stable::Tensor> cos_sin_cache);
void fused_kimi_k3_mla_key_concat_ds_mla_insert(
torch::stable::Tensor& q, torch::stable::Tensor const& k_nope,
torch::stable::Tensor const& k_pe, torch::stable::Tensor const& kv_c_normed,
torch::stable::Tensor& k_out, torch::stable::Tensor& k_cache,
torch::stable::Tensor const& slot_mapping, int64_t cache_block_size,
std::optional<torch::stable::Tensor> position_ids,
std::optional<torch::stable::Tensor> cos_sin_cache);
void fused_kimi_k3_mla_kv_concat(torch::stable::Tensor const& k_nope,
torch::stable::Tensor const& k_pe,
torch::stable::Tensor& k_out);
void fused_kimi_k3_mla_kv_concat_quant_fp8(torch::stable::Tensor const& k_nope,
torch::stable::Tensor const& k_pe,
torch::stable::Tensor const& v,
torch::stable::Tensor& k_fp8,
torch::stable::Tensor& v_fp8);
void fused_kimi_k3_mla_qkv_quant_kv_cache_fp8_insert(
torch::stable::Tensor const& q, torch::stable::Tensor const& k_nope,
torch::stable::Tensor const& k_pe, torch::stable::Tensor const& kv_c_normed,
torch::stable::Tensor const& v, torch::stable::Tensor& q_fp8,
torch::stable::Tensor& k_fp8, torch::stable::Tensor& v_fp8,
torch::stable::Tensor& k_cache, torch::stable::Tensor const& slot_mapping,
torch::stable::Tensor const& q_scale_inv,
torch::stable::Tensor const& k_scale_inv,
torch::stable::Tensor const& v_scale_inv,
torch::stable::Tensor const& cache_scale_inv, int64_t cache_block_size,
std::optional<torch::stable::Tensor> position_ids,
std::optional<torch::stable::Tensor> cos_sin_cache);
void fused_kimi_k3_mla_decode_q_concat_kv_cache_insert(
torch::stable::Tensor const& ql_nope, torch::stable::Tensor const& q_pe,
torch::stable::Tensor const& kv_c_normed, torch::stable::Tensor const& k_pe,
torch::stable::Tensor& mqa_q, torch::stable::Tensor& k_cache,
torch::stable::Tensor const& slot_mapping, int64_t cache_block_size,
std::optional<torch::stable::Tensor> position_ids,
std::optional<torch::stable::Tensor> cos_sin_cache);
void fused_kimi_k3_mla_decode_q_concat_kv_cache_fp8_insert(
torch::stable::Tensor const& ql_nope, torch::stable::Tensor const& q_pe,
torch::stable::Tensor const& kv_c_normed, torch::stable::Tensor const& k_pe,
torch::stable::Tensor& mqa_q, torch::stable::Tensor& k_cache,
torch::stable::Tensor const& slot_mapping,
torch::stable::Tensor const& q_scale_inv,
torch::stable::Tensor const& cache_scale_inv, int64_t cache_block_size,
std::optional<torch::stable::Tensor> position_ids,
std::optional<torch::stable::Tensor> cos_sin_cache);
void fused_kimi_k3_mla_decode_q_concat_ds_mla_insert(
torch::stable::Tensor const& ql_nope, torch::stable::Tensor const& q_pe,
torch::stable::Tensor const& kv_c_normed, torch::stable::Tensor const& k_pe,
torch::stable::Tensor& mqa_q, torch::stable::Tensor& k_cache,
torch::stable::Tensor const& slot_mapping, int64_t cache_block_size,
std::optional<torch::stable::Tensor> position_ids,
std::optional<torch::stable::Tensor> cos_sin_cache);
void fused_deepseek_v4_qnorm_rope_kv_rope_full_cache_fp8_insert(
torch::stable::Tensor const& q, torch::stable::Tensor const& kv,
torch::stable::Tensor& q_fp8, torch::stable::Tensor& k_cache,
torch::stable::Tensor const& slot_mapping,
torch::stable::Tensor const& position_ids,
torch::stable::Tensor const& cos_sin_cache,
torch::stable::Tensor const& fp8_scale,
torch::stable::Tensor const& q_fp8_scale_inv, double eps,
int64_t cache_block_size, bool apply_q_norm);
#ifndef USE_ROCM
std::tuple<torch::stable::Tensor, torch::stable::Tensor>
minimax_allreduce_rms_qk(torch::stable::Tensor qkv,
torch::stable::Tensor const& norm_weight_q,
torch::stable::Tensor const& norm_weight_k,
torch::stable::Tensor workspace, int64_t const q_size,
int64_t const kv_size, int64_t const rank,
int64_t const nranks, double const eps);
#endif
#ifndef USE_ROCM
namespace vllm {
void fused_qknorm_rope_cache(
torch::stable::Tensor& q_out, torch::stable::Tensor& k_out,
torch::stable::Tensor& v_out, torch::stable::Tensor& qkv,
torch::stable::Tensor& q_weight, torch::stable::Tensor& k_weight,
torch::stable::Tensor& cos, torch::stable::Tensor& sin,
torch::stable::Tensor& positions, torch::stable::Tensor& slot_mapping,
torch::stable::Tensor& k_cache, torch::stable::Tensor& v_cache,
int64_t head_dim, int64_t num_q_heads, int64_t num_kv_heads,
int64_t rotary_pairs, double epsilon, double norm_weight_bias,
const std::string& kv_cache_dtype,
std::optional<torch::stable::Tensor> k_scale,
std::optional<torch::stable::Tensor> v_scale);
} // namespace vllm
#endif
void fused_muse_spark_qknorm_rope_kv_insert(
torch::stable::Tensor& qkv, torch::stable::Tensor const& positions,
int64_t num_heads, int64_t num_kv_heads, int64_t head_dim, double eps,
std::optional<torch::stable::Tensor> cos_sin_cache, int64_t rotary_dim,
std::optional<torch::stable::Tensor> slot_mapping,
std::optional<torch::stable::Tensor> kv_cache, int64_t block_size,
const std::string& kv_cache_dtype,
std::optional<torch::stable::Tensor> k_scale,
std::optional<torch::stable::Tensor> v_scale,
std::optional<torch::stable::Tensor> q_fp8_out,
std::optional<torch::stable::Tensor> q_scale);
// Horizontally-fused MiniMax-M3 QK-norm + partial NeoX RoPE (+ optional KV /
// index-cache insert). Dense layer: norm+RoPE only; sparse layer: also packs
// the index branch and scatters k/v/index_k into their paged caches.
void fused_minimax_m3_qknorm_rope_kv_insert(
torch::stable::Tensor& qkv, torch::stable::Tensor const& q_norm_weight,
torch::stable::Tensor const& k_norm_weight,
torch::stable::Tensor const& cos_sin_cache,
torch::stable::Tensor const& positions, int64_t num_heads,
int64_t num_kv_heads, int64_t rotary_dim, double eps,
std::optional<torch::stable::Tensor> index_q_norm_weight,
std::optional<torch::stable::Tensor> index_k_norm_weight,
int64_t num_index_heads, std::optional<torch::stable::Tensor> slot_mapping,
std::optional<torch::stable::Tensor> index_slot_mapping,
std::optional<torch::stable::Tensor> kv_cache,
std::optional<torch::stable::Tensor> index_cache, int64_t block_size,
std::optional<torch::stable::Tensor> q_out,
std::optional<torch::stable::Tensor> index_q_out,
const std::string& kv_cache_dtype, bool skip_index_branch,
std::optional<torch::stable::Tensor> q_fp8_out, double q_fp8_scale,
std::optional<torch::stable::Tensor> kv_k_scale,
std::optional<torch::stable::Tensor> kv_v_scale);
#ifdef VLLM_ENABLE_FUSED_KDA_DECODE
void fused_kda_decode(
torch::stable::Tensor const& x, torch::stable::Tensor const& weight,
@@ -475,18 +314,6 @@ void fused_kda_chunk(
std::optional<torch::stable::Tensor> has_initial_state);
#endif
#ifdef VLLM_ENABLE_KIMI_K3_ATTN_RES
void kimi_k3_attn_res(torch::stable::Tensor& prefix,
std::optional<torch::stable::Tensor> delta,
torch::stable::Tensor& blocks,
torch::stable::Tensor const& norm_weight,
torch::stable::Tensor const& qk_weight,
std::optional<torch::stable::Tensor> output_norm_weight,
torch::stable::Tensor& output, int64_t num_blocks,
int64_t block_write_idx, double eps,
double output_norm_eps);
#endif
// Sampler kernels (shared CUDA/ROCm)
void apply_repetition_penalties_(
torch::stable::Tensor& logits, const torch::stable::Tensor& prompt_mask,
@@ -575,10 +402,6 @@ std::tuple<int64_t, torch::stable::Tensor> allocate_shared_buffer_and_handle(
int64_t open_mem_handle(torch::stable::Tensor& mem_handle);
void free_shared_buffer(int64_t buffer);
void muse_spark_mul_tanh_soft_cap(torch::stable::Tensor& out,
torch::stable::Tensor& input,
double multiplier, double soft_cap_temp);
// Activation kernels (shared CUDA/ROCm)
void silu_and_mul(torch::stable::Tensor& out, torch::stable::Tensor& input);
void silu_and_mul_clamp(torch::stable::Tensor& out,