llama + spec: MTP Support (#22673)

* spec: support MTP

* fix batch size

* rename files

* cont : simplify (#7)

* MTP: clean-up (#9)

* MTP: clean-up

* review: use llama_context_type instead of llama_graph_type

* review: remove llama_model_has_mtp

* review: fix convert issues

* convert: fix pycheck

* review: formatting

* use `mtp-` for identifying mtp models

* convert: fix mtp conversion

* mtp -> draft-mtp

* remove unused llama_arch

* add need_embd in speculative

* llama: allow partial seq_rm for GDN models for speculative decoding

Currently speculative checkpoint needs to restart from a checkpoint
after some draft tokens are not accepted, this leads to some wastage in
running the target again. This PR adds the ability to rollback upto
`draft_max` by storing the GDN intermediates.

* fix pending state

* vulkan: add GDN partial rollback

* meta: extend check to axis 1

* metal: add GDN partial rollback

Extend the gated delta net kernel to store intermediate states for
partial rollback support on the Metal backend.

- Add K (snapshot slot count) as a function constant
- Read input state from slot 0 of the 3D state tensor
- Write intermediate states to different slots during token loop
- For K=1, maintain backward-compatible single-slot behavior

Ref: https://github.com/ggml-org/llama.cpp/commit/8c05923630110223669f069af2000e9cf10c02bc

Assisted-by: llama.cpp:local pi

* delta_net_base: use ggml_pad instead of new_tensor

* review: add need_rs_seq

* review: rename part_bounded to n_rs

* review: deslop comments

* review: rename, add asserts

* server : adjust checkpoint logic (#11)

* server : adjust checkpoint logic

* cont : rm asserts

* server-context: fix early exit

* spec : fix compatibility with n-gram and add TODOs (#13)

* metal : cleanup

* llama : fix faulty bitwise check in recurrent memory

* server : disable RS-based MTP in combination with other spec types

* spec : add TODOs

* cont : fix comment

* cont : update comment

* common : fix logic for ngram + mtp compat

* llama-memory: enable checkpointing with partial rollback

* cont: add test-case for loading into a dirty ctx

* llama-memory-recurrent: clear rs_idx in clear

* download: fix mtp path

* llama-arch: fix enorm op

* docs: update docs

* conversion: fix type annotations

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

src/models/qwen3next.cpp

@@ -378,8 +378,6 @@ ggml_tensor * llama_model_qwen3next::graph::build_layer_attn_linear(
     const int64_t head_v_dim   = d_inner / num_v_heads;
     const int64_t n_seq_tokens = ubatch.n_seq_tokens;
 
-    const auto kv_head = mctx_cur->get_head();
-
     GGML_ASSERT(n_seqs != 0);
     GGML_ASSERT(ubatch.equal_seqs());
     GGML_ASSERT(ubatch.n_tokens == n_seq_tokens * n_seqs);
@@ -429,41 +427,14 @@ ggml_tensor * llama_model_qwen3next::graph::build_layer_attn_linear(
     beta = ggml_reshape_4d(ctx0, beta, 1, num_v_heads, n_seq_tokens, n_seqs);
     gate = ggml_reshape_4d(ctx0, gate, 1, num_v_heads, n_seq_tokens, n_seqs);
 
-    // Get convolution states from cache
     ggml_tensor * conv_states_all = mctx_cur->get_r_l(il);
     ggml_tensor * ssm_states_all  = mctx_cur->get_s_l(il);
 
-    // Build the convolution states tensor
-    ggml_tensor * conv_states = build_rs(inp, conv_states_all, hparams.n_embd_r(), n_seqs);
-    cb(conv_states, "conv_states", il);
-
-    // Calculate convolution kernel size
     ggml_tensor * conv_kernel      = model.layers[il].ssm_conv1d;
     const int64_t conv_kernel_size = conv_kernel->ne[0];
     const int64_t conv_channels    = d_inner + 2 * hparams.ssm_n_group * hparams.ssm_d_state;
 
-    conv_states = ggml_reshape_3d(ctx0, conv_states, conv_kernel_size - 1, conv_channels, n_seqs);
-    cb(conv_states, "conv_states_reshaped", il);
-
-    qkv_mixed = ggml_transpose(ctx0, qkv_mixed);
-    cb(qkv_mixed, "qkv_mixed_transposed", il);
-
-    ggml_tensor * conv_input = ggml_concat(ctx0, conv_states, qkv_mixed, 0);
-    cb(conv_input, "conv_input", il);
-
-    // Update convolution state cache
-    // Extract the last (conv_kernel_size - 1) states from conv_input
-    ggml_tensor * last_conv_states =
-        ggml_view_3d(ctx0, conv_input, conv_kernel_size - 1, conv_channels, n_seqs, conv_input->nb[1],
-                     conv_input->nb[2], (conv_input->ne[0] - conv_states->ne[0]) * ggml_element_size(conv_input));
-    cb(last_conv_states, "last_conv_states", il);
-
-    ggml_tensor * state_update_target =
-        ggml_view_2d(ctx0, conv_states_all, (conv_kernel_size - 1) * conv_channels, n_seqs, conv_states_all->nb[1],
-                     kv_head * (conv_kernel_size - 1) * conv_channels * ggml_element_size(conv_states_all));
-    cb(state_update_target, "state_update_target", il);
-
-    ggml_build_forward_expand(gf, ggml_cpy(ctx0, last_conv_states, state_update_target));
+    ggml_tensor * conv_input = build_conv_state(inp, conv_states_all, qkv_mixed, conv_kernel_size, conv_channels, il);
 
     ggml_tensor * state = build_rs(inp, ssm_states_all, hparams.n_embd_s(), n_seqs);
     state = ggml_reshape_4d(ctx0, state, head_v_dim, head_v_dim, num_v_heads, n_seqs);
@@ -540,18 +511,7 @@ ggml_tensor * llama_model_qwen3next::graph::build_layer_attn_linear(
     cb(k_conv, "k_conv_predelta", il);
     cb(v_conv, "v_conv_predelta", il);
 
-    auto attn_out = build_delta_net(q_conv, k_conv, v_conv, gate, beta, state, il);
-
-    ggml_tensor * output    = attn_out.first;
-    ggml_tensor * new_state = attn_out.second;
-    cb(output, "attn_output", il);
-    cb(new_state, "new_state", il);
-
-    // Update the recurrent states
-    ggml_build_forward_expand(gf,
-            ggml_cpy(ctx0, new_state,
-                ggml_view_2d(ctx0, ssm_states_all, hparams.n_embd_s(), n_seqs, ssm_states_all->nb[1],
-                    kv_head * hparams.n_embd_s() * ggml_element_size(ssm_states_all))));
+    ggml_tensor * output = build_recurrent_attn(inp, ssm_states_all, q_conv, k_conv, v_conv, gate, beta, state, il);
 
     // z: [head_dim, n_heads, n_tokens, n_seqs] -> [n_heads * n_tokens * n_seqs, head_dim]
     ggml_tensor * z_2d = ggml_reshape_4d(ctx0, z, head_v_dim, num_v_heads, n_seq_tokens, n_seqs);