Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74759.94347303537,
-      "required_pages_per_second_upper_estimate": 0.006480002422898259
+      "required_token_positions_per_second": 74881.75180674173,
+      "required_pages_per_second_upper_estimate": 0.006490560460543527
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9344.992934129421,
-      "required_pages_per_second_upper_estimate": 0.003727857401519635
+      "required_token_positions_per_second": 9360.218975842716,
+      "required_pages_per_second_upper_estimate": 0.003733931297208679
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18689.985868258842,
-      "required_pages_per_second_upper_estimate": 0.004323599298104842
+      "required_token_positions_per_second": 18720.43795168543,
+      "required_pages_per_second_upper_estimate": 0.00433064385166722
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28034.97880238826,
-      "required_pages_per_second_upper_estimate": 0.0017071831208992415
+      "required_token_positions_per_second": 28080.656927528147,
+      "required_pages_per_second_upper_estimate": 0.0017099646790656594
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28034.97880238826,
-      "required_pages_per_second_upper_estimate": 0.002158210273996165
+      "required_token_positions_per_second": 28080.656927528147,
+      "required_pages_per_second_upper_estimate": 0.0021617267025145766
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28034.97880238826,
-      "required_pages_per_second_upper_estimate": 0.3481006096873271
+      "required_token_positions_per_second": 28080.656927528147,
+      "required_pages_per_second_upper_estimate": 0.34866777912671376
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13151126917983902,
-    "producer_p95_seconds": 0.33250449399929494,
-    "gpu_update_seconds": 0.9861965723207433,
-    "producer_batches_per_second": 7.603911103865328,
-    "gpu_batches_per_second": 1.0139966291373068,
-    "producer_headroom": 7.498951066863626,
+    "producer_batch_seconds": 0.1191934512969226,
+    "producer_p95_seconds": 0.3040983800019603,
+    "gpu_update_seconds": 0.9845923502201259,
+    "producer_batches_per_second": 8.389722666129549,
+    "gpu_batches_per_second": 1.0156487603996003,
+    "producer_headroom": 8.260456757539552,
     "sustainable": true,
-    "buffer_seconds": 60.15799091156534,
+    "buffer_seconds": 60.06013336342768,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323349.0915038018,
+    "generation_bytes_per_second": 309779.68569474807,
     "upload_bytes_per_second": 5628047.752497826,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.38318785510579,
+    "estimated_full_shards_per_hour": 66.47150924808342,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -204,3 +204,4 @@
 {"step": 20400, "loss": 0.6723321489989758, "lr": 0.0003, "grad_norm": 0.13572539836841088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2984949151, "tokens_per_second": 148670.7590633348, "data_wait_s": 1.4277000445872545e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 20500, "loss": 0.6605964004993439, "lr": 0.0003, "grad_norm": 0.1411628422733053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2999634095, "tokens_per_second": 149195.88715567943, "data_wait_s": 1.0088999260915443e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 20600, "loss": 0.665947962552309, "lr": 0.0003, "grad_norm": 0.12847631823769204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3014375678, "tokens_per_second": 148735.51007187043, "data_wait_s": 1.3164000847609714e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 20700, "loss": 0.6963246762752533, "lr": 0.0003, "grad_norm": 0.14299341454153447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3029005928, "tokens_per_second": 148941.26953540367, "data_wait_s": 1.072499799192883e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e512e6f801063d301b72e2a55e9c4ad301c1c26df1ef4d45513dbe02660458a1
-size 433831
+oid sha256:a0e5a9ddba11c99ad0c1ff172087592cbfb2a3e6e7eb5cea113bf22566e62a2f
+size 435949