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": 76805.24292143255,
-      "required_pages_per_second_upper_estimate": 0.005313564918021202
+      "required_token_positions_per_second": 75816.20210286939,
+      "required_pages_per_second_upper_estimate": 0.0052451407793021295
     },
     {
       "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": 9600.65536517907,
-      "required_pages_per_second_upper_estimate": 0.004019593916918601
+      "required_token_positions_per_second": 9477.025262858673,
+      "required_pages_per_second_upper_estimate": 0.003967832574767161
     },
     {
       "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": 19201.31073035814,
-      "required_pages_per_second_upper_estimate": 0.003999126652517557
+      "required_token_positions_per_second": 18954.050525717346,
+      "required_pages_per_second_upper_estimate": 0.0039476288725809735
     },
     {
       "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": 28801.966095537206,
-      "required_pages_per_second_upper_estimate": 0.0017266338893716397
+      "required_token_positions_per_second": 28431.07578857602,
+      "required_pages_per_second_upper_estimate": 0.0017043995817860246
     },
     {
       "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": 28801.966095537206,
-      "required_pages_per_second_upper_estimate": 0.0022392261774678013
+      "required_token_positions_per_second": 28431.07578857602,
+      "required_pages_per_second_upper_estimate": 0.002210391087475621
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28801.966095537206,
-      "required_pages_per_second_upper_estimate": 0.3694406959317762
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28431.07578857602,
+      "required_pages_per_second_upper_estimate": 0.36468331330506304
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10124529882040179,
-    "producer_p95_seconds": 0.29676626099899295,
-    "gpu_update_seconds": 0.9599344679557827,
-    "producer_batches_per_second": 9.87700181293249,
-    "gpu_batches_per_second": 1.0417377783397426,
-    "producer_headroom": 9.48127448029565,
+    "producer_batch_seconds": 0.30739035670296744,
+    "producer_p95_seconds": 0.2877941600017948,
+    "gpu_update_seconds": 0.972457046845527,
+    "producer_batches_per_second": 3.253192490245568,
+    "gpu_batches_per_second": 1.0283230536956025,
+    "producer_headroom": 3.163589961884251,
     "sustainable": true,
-    "buffer_seconds": 60.47587148121431,
+    "buffer_seconds": 60.292336904422676,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349129.26220981515,
-    "upload_bytes_per_second": 4874240.16890082,
+    "generation_bytes_per_second": 360254.740781188,
+    "upload_bytes_per_second": 4876699.467468976,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.91501235695686,
+    "estimated_full_shards_per_hour": 77.30228107048731,
     "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": 1.216828146003536,
-      "bytes": 6096262,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.38459085499926,
       "bytes": 2407911,
@@ -562,6 +556,12 @@
       "bytes": 10383315,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.1864270160003798,
+      "bytes": 6161998,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -294,3 +294,4 @@
 {"step": 789400, "loss": 0.49407896399497986, "lr": 3e-05, "grad_norm": 0.2954638379966947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115545605874, "tokens_per_second": 152207.10726596077, "data_wait_s": 1.516799966339022e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 789500, "loss": 0.5112563371658325, "lr": 3e-05, "grad_norm": 0.2565419364300872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115560164229, "tokens_per_second": 150897.98640654597, "data_wait_s": 1.8455000827088952e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 789600, "loss": 0.507436215877533, "lr": 3e-05, "grad_norm": 0.4693924285816398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115574807096, "tokens_per_second": 152176.725787555, "data_wait_s": 1.5755002095829695e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 789700, "loss": 0.550376832485199, "lr": 3e-05, "grad_norm": 0.25415686008508515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115589472863, "tokens_per_second": 152485.19198147653, "data_wait_s": 1.161099862656556e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:50a16d70e491a8fc496a10840d33aa17bfe8138872065c7d158db692f1e8f072
-size 631008
+oid sha256:8d04c2415431a804c3e38ca596356f48ff8cd4128cd37c58a03a4b8bc4645401
+size 633126