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": 76722.54344231935,
-      "required_pages_per_second_upper_estimate": 0.005307843576166923
+      "required_token_positions_per_second": 76933.81764657935,
+      "required_pages_per_second_upper_estimate": 0.005322460015841323
     },
     {
       "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": 9590.317930289919,
-      "required_pages_per_second_upper_estimate": 0.004015265848800701
+      "required_token_positions_per_second": 9616.727205822419,
+      "required_pages_per_second_upper_estimate": 0.004026322860977774
     },
     {
       "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": 19180.635860579838,
-      "required_pages_per_second_upper_estimate": 0.003994820622375718
+      "required_token_positions_per_second": 19233.454411644838,
+      "required_pages_per_second_upper_estimate": 0.0040058213336436966
     },
     {
       "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": 28770.953790869757,
-      "required_pages_per_second_upper_estimate": 0.0017247747490598735
+      "required_token_positions_per_second": 28850.181617467257,
+      "required_pages_per_second_upper_estimate": 0.001729524336290507
     },
     {
       "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": 28770.953790869757,
-      "required_pages_per_second_upper_estimate": 0.002236815107188619
+      "required_token_positions_per_second": 28850.181617467257,
+      "required_pages_per_second_upper_estimate": 0.0022429747222201947
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28770.953790869757,
-      "required_pages_per_second_upper_estimate": 0.3690429033859206
+      "required_token_positions_per_second": 28850.181617467257,
+      "required_pages_per_second_upper_estimate": 0.3700591528773009
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.42486412724210254,
-    "producer_p95_seconds": 0.20935687199926178,
-    "gpu_update_seconds": 0.9609691844409372,
-    "producer_batches_per_second": 2.353693653759958,
-    "gpu_batches_per_second": 1.0406160948665277,
-    "producer_headroom": 2.261827070877516,
+    "producer_batch_seconds": 0.09049125033591565,
+    "producer_p95_seconds": 0.20480819899967173,
+    "gpu_update_seconds": 0.9583301889254174,
+    "producer_batches_per_second": 11.050792162644079,
+    "gpu_batches_per_second": 1.0434816846595507,
+    "producer_headroom": 10.590307741002222,
     "sustainable": true,
-    "buffer_seconds": 60.541058619779044,
+    "buffer_seconds": 60.3748019023013,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380174.84375796467,
-    "upload_bytes_per_second": 5573126.438450026,
+    "generation_bytes_per_second": 379428.29118526186,
+    "upload_bytes_per_second": 5657112.467204824,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.57667145303921,
+    "estimated_full_shards_per_hour": 81.41647864979164,
     "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": 0.8392939139998816,
-      "bytes": 2549967,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9600205120000282,
       "bytes": 6407325,
@@ -562,6 +556,12 @@
       "bytes": 10697456,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.3169437420001486,
+      "bytes": 11722554,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 568100, "loss": 0.5821261405944824, "lr": 0.0003, "grad_norm": 0.19712639032358825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83152926119, "tokens_per_second": 151624.50268147435, "data_wait_s": 1.1381998774595559e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 568200, "loss": 0.5839268565177917, "lr": 0.0003, "grad_norm": 0.16720505818914366, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83167589476, "tokens_per_second": 152423.64082107524, "data_wait_s": 1.2625001545529813e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 568300, "loss": 0.5627555847167969, "lr": 0.0003, "grad_norm": 0.18035638580757987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83182231142, "tokens_per_second": 152265.45215841712, "data_wait_s": 1.2732998584397137e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 568400, "loss": 0.5631968975067139, "lr": 0.0003, "grad_norm": 0.13816458641959578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83196871354, "tokens_per_second": 152554.49916727495, "data_wait_s": 1.3218999811215326e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c1c90d519c4f095af222d10692a959d699fe2b8f10864797524ff9cf530e70d
-size 238947
+oid sha256:6a511ccf4007c7d54a52b20b9b143b47f27ac0311b02efa8d4cf427649c1d318
+size 241065