server : disable context shift by default

ggml-ci

Author: ggerganov

common/arg.cpp

@@ -1530,6 +1530,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.ctx_shift = false;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_PERPLEXITY}).set_env("LLAMA_ARG_NO_CONTEXT_SHIFT"));
+    add_opt(common_arg(
+        {"--context-shift"},
+        string_format("enables context shift on infinite text generation (default: %s)", params.ctx_shift ? "disabled" : "enabled"),
+        [](common_params & params) {
+            params.ctx_shift = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_PERPLEXITY}).set_env("LLAMA_ARG_CONTEXT_SHIFT"));
     add_opt(common_arg(
         {"--chunks"}, "N",
         string_format("max number of chunks to process (default: %d, -1 = all)", params.n_chunks),

common/common.h

@@ -372,7 +372,7 @@ struct common_params {
     bool cont_batching     = true;  // insert new sequences for decoding on-the-fly
     bool flash_attn        = false; // flash attention
     bool no_perf           = false; // disable performance metrics
-    bool ctx_shift         = true;  // context shift on inifinite text generation
+    bool ctx_shift         = false;  // context shift on inifinite text generation
     bool swa_full          = false; // use full-size SWA cache (https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
     bool kv_unified        = false; // enable unified KV cache
 

tools/server/tests/unit/test_completion.py

@@ -229,7 +229,7 @@ def test_nocache_long_input_prompt():
         "temperature": 1.0,
         "cache_prompt": False,
     })
-    assert res.status_code == 200
+    assert res.status_code == 400
 
 
 def test_completion_with_tokens_input():

tools/server/tests/unit/test_ctx_shift.py

@@ -25,7 +25,9 @@ def test_ctx_shift_enabled():
     # the prompt is truncated to keep the last 109 tokens
     # 64 tokens are generated thanks to shifting the context when it gets full
     global server
+    server.enable_ctx_shift = True
     server.start()
+    server.enable_ctx_shift = False
     res = server.make_request("POST", "/completion", data={
         "n_predict": 64,
         "prompt": LONG_TEXT,
@@ -42,7 +44,6 @@ def test_ctx_shift_enabled():
 ])
 def test_ctx_shift_disabled_short_prompt(n_predict: int, n_token_output: int, truncated: bool):
     global server
-    server.disable_ctx_shift = True
     server.n_predict = -1
     server.start()
     res = server.make_request("POST", "/completion", data={
@@ -56,7 +57,6 @@ def test_ctx_shift_disabled_short_prompt(n_predict: int, n_token_output: int, tr
 
 def test_ctx_shift_disabled_long_prompt():
     global server
-    server.disable_ctx_shift = True
     server.start()
     res = server.make_request("POST", "/completion", data={
         "n_predict": 64,
@@ -68,7 +68,6 @@ def test_ctx_shift_disabled_long_prompt():
 
 def test_ctx_shift_disabled_stream():
     global server
-    server.disable_ctx_shift = True
     server.start()
     res = server.make_stream_request("POST", "/v1/completions", data={
         "n_predict": 256,

tools/server/tests/unit/test_speculative.py

@@ -91,7 +91,9 @@ def test_slot_ctx_not_exceeded():
 def test_with_ctx_shift():
     global server
     server.n_ctx = 64
+    server.enable_ctx_shift = True
     server.start()
+    server.enable_ctx_shift = False
     res = server.make_request("POST", "/completion", data={
         "prompt": "Hello " * 56,
         "temperature": 0.0,

tools/server/tests/unit/test_tool_call.py

@@ -22,6 +22,8 @@ def create_server():
     server.model_alias = "tinyllama-2-tool-call"
     server.server_port = 8081
     server.n_slots = 1
+    server.n_ctx = 8192
+    server.n_batch = 2048
 
 class CompletionMode(Enum):
     NORMAL = "normal"

tools/server/tests/utils.py

@@ -79,7 +79,7 @@ class ServerProcess:
     draft: int | None = None
     api_key: str | None = None
     lora_files: List[str] | None = None
-    disable_ctx_shift: int | None = False
+    enable_ctx_shift: int | None = False
     draft_min: int | None = None
     draft_max: int | None = None
     no_webui: bool | None = None
@@ -178,8 +178,8 @@ def start(self, timeout_seconds: int | None = DEFAULT_HTTP_TIMEOUT) -> None:
         if self.lora_files:
             for lora_file in self.lora_files:
                 server_args.extend(["--lora", lora_file])
-        if self.disable_ctx_shift:
-            server_args.extend(["--no-context-shift"])
+        if self.enable_ctx_shift:
+            server_args.append("--context-shift")
         if self.api_key:
             server_args.extend(["--api-key", self.api_key])
         if self.draft_max:

tools/tts/tts.cpp

@@ -581,7 +581,6 @@ int main(int argc, char ** argv) {
 
     params.model = params.vocoder.model;
     params.embedding = true;
-    params.ctx_shift = false; // silence warning
     params.n_ubatch = params.n_batch;
 
     common_init_result llama_init_cts = common_init_from_params(params);