recurrent : rework graph inputs + add TODOs by ggerganov · Pull Request #2 · gabe-l-hart/llama.cpp

ggerganov · 2025-06-18T06:38:23Z

The diff is a bit confusing because I also moved the definitions in llama-graph.cpp to follow the order of the declarations in llama-graph.h.

ggml-ci

ggerganov · 2025-06-18T06:38:58Z

src/llama-graph.cpp

+    const int64_t n_kv = kv_state->get_state_recurrent()->get_n_kv();
+
+    if (s_copy) {
+        GGML_ASSERT(ggml_backend_buffer_is_host(s_copy->buffer));
+        int32_t * data = (int32_t *) s_copy->data;
+
+        // assuming copy destinations ALWAYS happen ONLY on the cells between head and head+n
+        for (uint32_t i = 0; i < n_kv; ++i) {
+            data[i] = kv_state->get_state_recurrent()->s_copy(i);
+        }
+    }


Move to a set_input_... method similar to set_input_kq_mask() above.

gabe-l-hart

Thank you for all the updates here! I'll merge and look at a couple more tweaks on my fork, then 🤞 we should be close on the hybrid cache.

gabe-l-hart · 2025-06-18T13:10:01Z

src/llama-graph.cpp

-
 ggml_tensor * llm_graph_context::build_attn(
-        llm_graph_input_attn_kv_hybrid_recurrent * inp,
+        llm_graph_input_attn_kv_unified_iswa * inp,


Ah, yep, funky diff, but this is just the reorder, right?

gabe-l-hart · 2025-06-18T13:14:37Z

src/llama-graph.cpp

+    return (llm_graph_input_attn_kv_unified_iswa *) res->add_input(std::move(inp));
+}
+
+ggml_tensor * llm_graph_context::build_rs(


I think I'm parsing the diff correctly that this now fully removes build_recurrent_state and makes this the input-agnostic implementation of build_rs instead. I think that makes sense since they're ultimately different signatures to the same functionality.

Yes, exactly. I think this function has to be eventually moved to the recurrent cache implementation. See the TODO that I left in the declaration.

gabe-l-hart · 2025-06-18T13:20:46Z

src/llama-kv-cache-hybrid-recurrent.cpp

+        llama_kv_cache_hybrid_recurrent * kv,
+        llama_context * lctx,
+        bool optimize) : status(LLAMA_MEMORY_STATUS_SUCCESS) {
+    state_attn      = kv->get_kv_attn     ()->init_update(lctx, optimize);


Nice, this makes more sense to encapsulate the creation of the child states here instead of in the init methods

gabe-l-hart · 2025-06-18T13:25:16Z

src/llama-kv-cache-hybrid-recurrent.cpp

-      state_attn(new llama_kv_cache_unified_state(kv->get_kv_attn())),
-      state_recurrent(new llama_kv_cache_recurrent_state(status, kv->get_kv_recurrent())) {}
+    : status(LLAMA_MEMORY_STATUS_SUCCESS) {
+    state_attn      = kv->get_kv_attn     ()->init_full();


I think these can be moved back into the initializer list and then we can hold status/state_attn/state_recurrent as const again. I'll try that out after merging.

gabe-l-hart · 2025-06-18T13:26:10Z

src/llama-kv-cache-hybrid-recurrent.h

    const llama_kv_cache_recurrent_state * get_state_recurrent() const;

 private:
-    const llama_memory_status status;


I think we can restore these const qualifiers if we tweak the "init update" constructor signature a little. I'll give it a shot once this is merged to my fork.

I changed this to be consistent with the iswa implementation. If you move back, then also update the iswa cache respectively, to keep the consistency. Can be done in small follow-up PR too.

Ah, makes sense. I'll see if it's easy to do for both and leave it if not

Ok, it was pretty simple, so I just pushed that to the branch

…gml-org#16038) Initalizing RESERVED_NAME in is_reserved_name() is not thread safe and leads to corrupted memory when used from multiple threads as can be seen in the asan trace below. This fixes the initialization to make it thread-safe. #0 0x000100abd018 in std::__1::pair<std::__1::__hash_iterator<std::__1::__hash_node<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, void*>*>, bool> std::__1::__hash_table<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, std::__1::hash<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>>, std::__1::equal_to<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>>, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>>>::__emplace_unique_key_args<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&>(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&) __hash_table:1565 #1 0x000100ab0320 in SchemaConverter::visit(nlohmann::json_abi_v3_12_0::basic_json<nlohmann::json_abi_v3_12_0::ordered_map, std::__1::vector, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool, long long, unsigned long long, double, std::__1::allocator, nlohmann::json_abi_v3_12_0::adl_serializer, std::__1::vector<unsigned char, std::__1::allocator<unsigned char>>, void> const&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&) json-schema-to-grammar.cpp:802 #2 0x000100aafc48 in std::__1::__function::__func<build_grammar(std::__1::function<void (common_grammar_builder const&)> const&, common_grammar_options const&)::$_2, std::__1::allocator<build_grammar(std::__1::function<void (common_grammar_builder const&)> const&, common_grammar_options const&)::$_2>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> (std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&, nlohmann::json_abi_v3_12_0::basic_json<nlohmann::json_abi_v3_12_0::ordered_map, std::__1::vector, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool, long long, unsigned long long, double, std::__1::allocator, nlohmann::json_abi_v3_12_0::adl_serializer, std::__1::vector<unsigned char, std::__1::allocator<unsigned char>>, void> const&)>::operator()(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&, nlohmann::json_abi_v3_12_0::basic_json<nlohmann::json_abi_v3_12_0::ordered_map, std::__1::vector, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool, long long, unsigned long long, double, std::__1::allocator, nlohmann::json_abi_v3_12_0::adl_serializer, std::__1::vector<unsigned char, std::__1::allocator<unsigned char>>, void> const&) function.h:319 #3 0x000100a2c938 in std::__1::__function::__func<common_chat_params_init_llama_3_x(minja::chat_template const&, templates_params const&, bool)::$_0::operator()(common_grammar_builder const&) const::'lambda'(nlohmann::json_abi_v3_12_0::basic_json<nlohmann::json_abi_v3_12_0::ordered_map, std::__1::vector, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool, long long, unsigned long long, double, std::__1::allocator, nlohmann::json_abi_v3_12_0::adl_serializer, std::__1::vector<unsigned char, std::__1::allocator<unsigned char>>, void> const&), std::__1::allocator<common_chat_params_init_llama_3_x(minja::chat_template const&, templates_params const&, bool)::$_0::operator()(common_grammar_builder const&) const::'lambda'(nlohmann::json_abi_v3_12_0::basic_json<nlohmann::json_abi_v3_12_0::ordered_map, std::__1::vector, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool, long long, unsigned long long, double, std::__1::allocator, nlohmann::json_abi_v3_12_0::adl_serializer, std::__1::vector<unsigned char, std::__1::allocator<unsigned char>>, void> const&)>, void (nlohmann::json_abi_v3_12_0::basic_json<nlohmann::json_abi_v3_12_0::ordered_map, std::__1::vector, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool, long long, unsigned long long, double, std::__1::allocator, nlohmann::json_abi_v3_12_0::adl_serializer, std::__1::vector<unsigned char, std::__1::allocator<unsigned char>>, void> const&)>::operator()(nlohmann::json_abi_v3_12_0::basic_json<nlohmann::json_abi_v3_12_0::ordered_map, std::__1::vector, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool, long long, unsigned long long, double, std::__1::allocator, nlohmann::json_abi_v3_12_0::adl_serializer, std::__1::vector<unsigned char, std::__1::allocator<unsigned char>>, void> const&) function.h:319 #4 0x000100a139f8 in foreach_function(nlohmann::json_abi_v3_12_0::basic_json<nlohmann::json_abi_v3_12_0::ordered_map, std::__1::vector, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool, long long, unsigned long long, double, std::__1::allocator, nlohmann::json_abi_v3_12_0::adl_serializer, std::__1::vector<unsigned char, std::__1::allocator<unsigned char>>, void> const&, std::__1::function<void (nlohmann::json_abi_v3_12_0::basic_json<nlohmann::json_abi_v3_12_0::ordered_map, std::__1::vector, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool, long long, unsigned long long, double, std::__1::allocator, nlohmann::json_abi_v3_12_0::adl_serializer, std::__1::vector<unsigned char, std::__1::allocator<unsigned char>>, void> const&)> const&) chat.cpp:762 ggml-org#5 0x000100a2a7f4 in std::__1::__function::__func<common_chat_params_init_llama_3_x(minja::chat_template const&, templates_params const&, bool)::$_0, std::__1::allocator<common_chat_params_init_llama_3_x(minja::chat_template const&, templates_params const&, bool)::$_0>, void (common_grammar_builder const&)>::operator()(common_grammar_builder const&) function.h:319 ggml-org#6 0x000100aa98f4 in build_grammar(std::__1::function<void (common_grammar_builder const&)> const&, common_grammar_options const&) json-schema-to-grammar.cpp:982 ggml-org#7 0x0001009c9314 in common_chat_params_init_llama_3_x(minja::chat_template const&, templates_params const&, bool) chat.cpp:1110 ggml-org#8 0x0001009b8afc in common_chat_templates_apply_jinja(common_chat_templates const*, common_chat_templates_inputs const&) chat.cpp:1992 ggml-org#9 0x0001009b533c in common_chat_templates_apply(common_chat_templates const*, common_chat_templates_inputs const&) chat.cpp:2074 ggml-org#10 0x000100810120 in llamacpp_apply_chat_template+0x724 (predict_oai-98384e17fb94e863:arm64+0x100090120) ... ==45482==Register values: x[0] = 0x00006020004147f8 x[1] = 0x00006080000013c8 x[2] = 0x0000000000000000 x[3] = 0x0000604006289738 x[4] = 0x0000000000000002 x[5] = 0x0000000000000001 x[6] = 0x04034000004b4000 x[7] = 0x0000000000000001 x[8] = 0xbebebebebebebebe x[9] = 0x17d7d7d7d7d7d7d7 x[10] = 0x00000c04000828ff x[11] = 0x0000000000000001 x[12] = 0x000000002018d383 x[13] = 0x0000000000000000 x[14] = 0xfa0000000000fafa x[15] = 0x000010700001ffff x[16] = 0x000000019dc012c0 x[17] = 0x00000001021284f8 x[18] = 0x0000000000000000 x[19] = 0x00000001700acdc0 x[20] = 0x0000000000000002 x[21] = 0x000000002018d384 x[22] = 0x16dd16fd2e731151 x[23] = 0x0000007000020000 x[24] = 0x0000000100c69c08 x[25] = 0x0000000100c69c20 x[26] = 0x00006080000013c7 x[27] = 0x0000000100c69c00 x[28] = 0x00000001700acd60 fp = 0x00000001700aceb0 lr = 0x0000000100abce30 sp = 0x00000001700acd60 AddressSanitizer can not provide additional info. SUMMARY: AddressSanitizer: SEGV __hash_table:1565 in std::__1::pair<std::__1::__hash_iterator<std::__1::__hash_node<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, void*>*>, bool> std::__1::__hash_table<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, std::__1::hash<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>>, std::__1::equal_to<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>>, std::__1::allocator<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>>>::__emplace_unique_key_args<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&>(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&) Thread T5 created by T0 here: #0 0x0001020b99d4 in pthread_create+0x5c (libclang_rt.asan_osx_dynamic.dylib:arm64e+0x359d4) #1 0x000100873910 in std::sys::pal::unix::thread::Thread::new::h77254fdd87a28e05+0x118 (predict_oai-98384e17fb94e863:arm64+0x1000f3910) #2 0x0001007c7a1c in test::run_test::haeb3c2bcd5ed6cf6+0x76c (predict_oai-98384e17fb94e863:arm64+0x100047a1c) #3 0x0001007aedb0 in test::console::run_tests_console::he9d142d704f3a986+0x149c (predict_oai-98384e17fb94e863:arm64+0x10002edb0) #4 0x0001007c5758 in test::test_main::hf86a5e20735245b9+0x118 (predict_oai-98384e17fb94e863:arm64+0x100045758) ggml-org#5 0x0001007c5da0 in test::test_main_static::h61ee9c8fd30abca0+0x54 (predict_oai-98384e17fb94e863:arm64+0x100045da0) ... ==45482==ABORTING

recurrent : rework graph inputs + add TODOs

59fee24

ggml-ci

ggerganov commented Jun 18, 2025

View reviewed changes

ggerganov mentioned this pull request Jun 18, 2025

Hybrid recurrent cache ggml-org/llama.cpp#13979

Merged

gabe-l-hart approved these changes Jun 18, 2025

View reviewed changes

gabe-l-hart merged commit c80e68c into gabe-l-hart:HybridRecurrentCache Jun 18, 2025
1 check passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

recurrent : rework graph inputs + add TODOs#2

recurrent : rework graph inputs + add TODOs#2
gabe-l-hart merged 1 commit intogabe-l-hart:HybridRecurrentCachefrom
ggml-org:gabe-l-hart/HybridRecurrentCache

ggerganov commented Jun 18, 2025 •

edited

Loading

Uh oh!

ggerganov Jun 18, 2025

Uh oh!

gabe-l-hart left a comment

Uh oh!

gabe-l-hart Jun 18, 2025

Uh oh!

ggerganov Jun 18, 2025

Uh oh!

gabe-l-hart Jun 18, 2025

Uh oh!

ggerganov Jun 18, 2025

Uh oh!

gabe-l-hart Jun 18, 2025

Uh oh!

gabe-l-hart Jun 18, 2025

Uh oh!

gabe-l-hart Jun 18, 2025

Uh oh!

ggerganov Jun 18, 2025

Uh oh!

gabe-l-hart Jun 18, 2025

Uh oh!

gabe-l-hart Jun 18, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

ggerganov commented Jun 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gabe-l-hart left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

ggerganov commented Jun 18, 2025 •

edited

Loading