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": 75749.55767851035,
-      "required_pages_per_second_upper_estimate": 0.006565779674074557
+      "required_token_positions_per_second": 75818.46909376918,
+      "required_pages_per_second_upper_estimate": 0.0065717527408948075
     },
     {
       "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": 9468.694709813793,
-      "required_pages_per_second_upper_estimate": 0.003777203889346495
+      "required_token_positions_per_second": 9477.308636721147,
+      "required_pages_per_second_upper_estimate": 0.0037806401135795226
     },
     {
       "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": 18937.389419627587,
-      "required_pages_per_second_upper_estimate": 0.00438083175556021
+      "required_token_positions_per_second": 18954.617273442294,
+      "required_pages_per_second_upper_estimate": 0.004384817116340371
     },
     {
       "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": 28406.08412944138,
-      "required_pages_per_second_upper_estimate": 0.0017297814882776005
+      "required_token_positions_per_second": 28431.92591016344,
+      "required_pages_per_second_upper_estimate": 0.0017313551171422288
     },
     {
       "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": 28406.08412944138,
-      "required_pages_per_second_upper_estimate": 0.002186778989358008
+      "required_token_positions_per_second": 28431.92591016344,
+      "required_pages_per_second_upper_estimate": 0.0021887683611726202
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28406.08412944138,
-      "required_pages_per_second_upper_estimate": 0.3527084958396933
+      "required_token_positions_per_second": 28431.92591016344,
+      "required_pages_per_second_upper_estimate": 0.35302936426938475
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.09738114971867162,
-    "producer_p95_seconds": 0.3122599630005425,
-    "gpu_update_seconds": 0.9733126140869355,
-    "producer_batches_per_second": 10.268927845778581,
-    "gpu_batches_per_second": 1.0274191308391702,
-    "producer_headroom": 9.994877005444874,
+    "producer_batch_seconds": 0.10982468016405278,
+    "producer_p95_seconds": 0.3197330740003963,
+    "gpu_update_seconds": 0.9724279701403128,
+    "producer_batches_per_second": 9.105421463611187,
+    "gpu_batches_per_second": 1.028353801727555,
+    "producer_headroom": 8.854366511131463,
     "sustainable": true,
-    "buffer_seconds": 60.34538207339,
+    "buffer_seconds": 60.29053414869939,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347624.86882644735,
-    "upload_bytes_per_second": 3953797.862974227,
+    "generation_bytes_per_second": 360141.4758567557,
+    "upload_bytes_per_second": 3956380.445076952,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.59220455737176,
+    "estimated_full_shards_per_hour": 77.27797705437663,
     "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": 0.888292913998157,
-      "bytes": 897869,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.0212596540004597,
       "bytes": 5845601,
@@ -562,6 +556,12 @@
       "bytes": 12178587,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 0.8894950190006057,
+      "bytes": 1180133,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 139200, "loss": 0.6169905085116625, "lr": 0.0003, "grad_norm": 0.12975897033810646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20371814263, "tokens_per_second": 149834.12120888062, "data_wait_s": 9.225001122104004e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 139300, "loss": 0.5964872129261494, "lr": 0.0003, "grad_norm": 0.1318150963052732, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20386459708, "tokens_per_second": 149490.33986902668, "data_wait_s": 1.4132998330751434e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 139400, "loss": 0.6256307102739811, "lr": 0.0003, "grad_norm": 0.14989689171879073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20401218446, "tokens_per_second": 150023.24210697704, "data_wait_s": 1.4225999620975927e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 139500, "loss": 0.6047249548137188, "lr": 0.0003, "grad_norm": 0.13372338313865564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20415982038, "tokens_per_second": 150755.5205487166, "data_wait_s": 1.6589998267591e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0d2d2794a3a842180f00110fbd6b9ef6c7bce998091bb037bdd6871a22e97b3b
-size 520089
+oid sha256:7875de7ee8d062f72c8db47461f972aba27ae16979b866d8b084d81af0f3f67e
+size 522207