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": 76594.57833129536,
-      "required_pages_per_second_upper_estimate": 0.005298990652866317
+      "required_token_positions_per_second": 76962.51468433492,
+      "required_pages_per_second_upper_estimate": 0.005324445343499555
     },
     {
       "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": 9574.32229141192,
-      "required_pages_per_second_upper_estimate": 0.004008568808829407
+      "required_token_positions_per_second": 9620.314335541865,
+      "required_pages_per_second_upper_estimate": 0.00402782471728352
     },
     {
       "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": 19148.64458282384,
-      "required_pages_per_second_upper_estimate": 0.003988157682885873
+      "required_token_positions_per_second": 19240.62867108373,
+      "required_pages_per_second_upper_estimate": 0.004007315542686874
     },
     {
       "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": 28722.966874235757,
-      "required_pages_per_second_upper_estimate": 0.0017218980066794442
+      "required_token_positions_per_second": 28860.943006625595,
+      "required_pages_per_second_upper_estimate": 0.0017301694651388597
     },
     {
       "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": 28722.966874235757,
-      "required_pages_per_second_upper_estimate": 0.002233084335492465
+      "required_token_positions_per_second": 28860.943006625595,
+      "required_pages_per_second_upper_estimate": 0.002243811372892907
     },
     {
       "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": 28722.966874235757,
-      "required_pages_per_second_upper_estimate": 0.3684273787436764
+      "required_token_positions_per_second": 28860.943006625595,
+      "required_pages_per_second_upper_estimate": 0.3701971884227446
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.0879893581092972,
-    "producer_p95_seconds": 0.33382803600034094,
-    "gpu_update_seconds": 0.962574657452953,
-    "producer_batches_per_second": 11.365010740933423,
-    "gpu_batches_per_second": 1.038880456967439,
-    "producer_headroom": 10.93967132090312,
+    "producer_batch_seconds": 0.1575207477346794,
+    "producer_p95_seconds": 0.31139828199957265,
+    "gpu_update_seconds": 0.9579728560377553,
+    "producer_batches_per_second": 6.348370067950372,
+    "gpu_batches_per_second": 1.0438709131447335,
+    "producer_headroom": 6.0815662051790165,
     "sustainable": true,
-    "buffer_seconds": 60.64220341953604,
+    "buffer_seconds": 60.35228993037858,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277404.1715783259,
-    "upload_bytes_per_second": 5472608.967769391,
+    "generation_bytes_per_second": 310001.39995414193,
+    "upload_bytes_per_second": 5561567.9521295475,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.524477582095464,
+    "estimated_full_shards_per_hour": 66.51908396690553,
     "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": 1.3192006099998252,
-      "bytes": 2683646,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.3172165699979814,
       "bytes": 5876805,
@@ -562,6 +556,12 @@
       "bytes": 2290704,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.766581726995355,
+      "bytes": 12101436,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -333,3 +333,4 @@
 {"step": 678300, "loss": 0.5567049384117126, "lr": 0.0001, "grad_norm": 0.19036875865297515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99285865302, "tokens_per_second": 152639.1467377215, "data_wait_s": 9.129998943535611e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 678400, "loss": 0.5780566334724426, "lr": 0.0001, "grad_norm": 0.21393389684835812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99300550481, "tokens_per_second": 153261.01173797937, "data_wait_s": 1.1306998203508556e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 678500, "loss": 0.5544663071632385, "lr": 0.0001, "grad_norm": 0.18759753303234136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99315125954, "tokens_per_second": 151424.7293657391, "data_wait_s": 1.0864998330362141e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 678600, "loss": 0.5879587531089783, "lr": 0.0001, "grad_norm": 0.19140846977270953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99329593942, "tokens_per_second": 150517.2255848945, "data_wait_s": 9.38000448513776e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63bf054853c82052ded2286f6502c031ec1fad9002f83fe295a5ba298aabd5fe
-size 714393
+oid sha256:5fd1a65d746a15332b8a17f1070cbb6a698de1655d6cc24791625930f45873a1
+size 716511