Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76097.9926286936,
-      "required_pages_per_second_upper_estimate": 0.00526463570172275
+      "required_token_positions_per_second": 75566.84662885501,
+      "required_pages_per_second_upper_estimate": 0.005227889789025402
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9512.2490785867,
-      "required_pages_per_second_upper_estimate": 0.003982580050855571
+      "required_token_positions_per_second": 9445.855828606876,
+      "required_pages_per_second_upper_estimate": 0.003954782583537737
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19024.4981571734,
-      "required_pages_per_second_upper_estimate": 0.0039623012563842945
+      "required_token_positions_per_second": 18891.711657213753,
+      "required_pages_per_second_upper_estimate": 0.003934645330258191
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28536.747235760096,
-      "required_pages_per_second_upper_estimate": 0.0017107344236937526
+      "required_token_positions_per_second": 28337.567485820633,
+      "required_pages_per_second_upper_estimate": 0.0016987939018147501
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28536.747235760096,
-      "required_pages_per_second_upper_estimate": 0.002218606577694608
+      "required_token_positions_per_second": 28337.567485820633,
+      "required_pages_per_second_upper_estimate": 0.0022031212282359455
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28536.747235760096,
-      "required_pages_per_second_upper_estimate": 0.3660387531683803
+      "required_token_positions_per_second": 28337.567485820633,
+      "required_pages_per_second_upper_estimate": 0.36348388919871005
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12064255759381126,
-    "producer_p95_seconds": 0.2690781480014266,
-    "gpu_update_seconds": 0.9688560427571652,
-    "producer_batches_per_second": 8.288948940944021,
-    "gpu_batches_per_second": 1.0321450823119247,
-    "producer_headroom": 8.03079826953922,
+    "producer_batch_seconds": 0.07543878170309881,
+    "producer_p95_seconds": 0.2577721269990434,
+    "gpu_update_seconds": 0.9756659605251695,
+    "producer_batches_per_second": 13.255781408767406,
+    "gpu_batches_per_second": 1.0249409536248781,
+    "producer_headroom": 12.933214700696736,
     "sustainable": true,
-    "buffer_seconds": 60.069074650944245,
+    "buffer_seconds": 60.49128955256051,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409311.35698327195,
-    "upload_bytes_per_second": 5309535.600740369,
+    "generation_bytes_per_second": 409357.0586202034,
+    "upload_bytes_per_second": 5256754.196238368,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.8286889278757,
+    "estimated_full_shards_per_hour": 87.83849543528153,
     "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": [
-    {
-      "seconds": 1.667742324003484,
-      "bytes": 13340513,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.2163241669986746,
       "bytes": 1010699,
@@ -562,6 +556,12 @@
       "bytes": 7121753,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.1241114900003595,
+      "bytes": 6180535,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -241,3 +241,4 @@
 {"step": 744100, "loss": 0.5697230100631714, "lr": 5e-05, "grad_norm": 0.24556399169410378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108915801769, "tokens_per_second": 149881.75720913263, "data_wait_s": 1.5187997632892802e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 744200, "loss": 0.5490539073944092, "lr": 5e-05, "grad_norm": 0.22234718924079527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108930354542, "tokens_per_second": 149898.92008542002, "data_wait_s": 1.2029999197693542e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 744300, "loss": 0.5325649976730347, "lr": 5e-05, "grad_norm": 0.24354972110028025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108945015345, "tokens_per_second": 150717.29578819821, "data_wait_s": 1.0838000889634714e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 744400, "loss": 0.5335682034492493, "lr": 5e-05, "grad_norm": 0.2286610199415738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108959682410, "tokens_per_second": 151428.75428831435, "data_wait_s": 1.5768000594107434e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:65a11ef5fd3d26890eec84cfed2eff6ec8b2c4037b49659922b9e0898d5df1fe
-size 517971
+oid sha256:a3de5ba05e0eedc2ff064ed11ba34f015464b56e5c6ffa731c19cc9dc29a538e
+size 520089