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": 73251.94770855844,
-      "required_pages_per_second_upper_estimate": 0.006349293172013722
+      "required_token_positions_per_second": 73513.31592061793,
+      "required_pages_per_second_upper_estimate": 0.006371947906203361
     },
     {
       "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": 9156.493463569805,
-      "required_pages_per_second_upper_estimate": 0.003652662144395167
+      "required_token_positions_per_second": 9189.16449007724,
+      "required_pages_per_second_upper_estimate": 0.0036656951053443596
     },
     {
       "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": 18312.98692713961,
-      "required_pages_per_second_upper_estimate": 0.00423638722803166
+      "required_token_positions_per_second": 18378.32898015448,
+      "required_pages_per_second_upper_estimate": 0.004251502962015794
     },
     {
       "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": 27469.480390709414,
-      "required_pages_per_second_upper_estimate": 0.0016727472345688685
+      "required_token_positions_per_second": 27567.493470231722,
+      "required_pages_per_second_upper_estimate": 0.0016787157168768909
     },
     {
       "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": 27469.480390709414,
-      "required_pages_per_second_upper_estimate": 0.0021146766408653337
+      "required_token_positions_per_second": 27567.493470231722,
+      "required_pages_per_second_upper_estimate": 0.00212222195904453
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27469.480390709414,
-      "required_pages_per_second_upper_estimate": 0.3410790120157122
+      "required_token_positions_per_second": 27567.493470231722,
+      "required_pages_per_second_upper_estimate": 0.3422960064347036
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.32714445918742285,
-    "producer_p95_seconds": 0.3968580220025615,
-    "gpu_update_seconds": 1.0064988345884753,
-    "producer_batches_per_second": 3.0567535897867506,
-    "gpu_batches_per_second": 0.9935431275574874,
-    "producer_headroom": 3.076618925744503,
+    "producer_batch_seconds": 0.1241730749608223,
+    "producer_p95_seconds": 0.4278223639994394,
+    "gpu_update_seconds": 1.0029203427527864,
+    "producer_batches_per_second": 8.05327564220753,
+    "gpu_batches_per_second": 0.9970881608156729,
+    "producer_headroom": 8.076793967365441,
     "sustainable": true,
-    "buffer_seconds": 60.38993007530852,
+    "buffer_seconds": 60.17522056516719,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458696.2961486484,
-    "upload_bytes_per_second": 5822085.459185429,
+    "generation_bytes_per_second": 460038.7097324806,
+    "upload_bytes_per_second": 5787306.752317857,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.42554725021924,
+    "estimated_full_shards_per_hour": 98.71359795552075,
     "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": 1.0841833719969145,
-      "bytes": 7541614,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8043565109983319,
       "bytes": 540912,
@@ -562,6 +556,12 @@
       "bytes": 10600974,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.208974320004927,
+      "bytes": 5767450,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -403,3 +403,4 @@
 {"step": 395300, "loss": 0.5941524654626846, "lr": 0.0003, "grad_norm": 0.1554837989482024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57860678821, "tokens_per_second": 148201.14242640784, "data_wait_s": 9.780000254977494e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 395400, "loss": 0.5709842573851347, "lr": 0.0003, "grad_norm": 0.1689879688406585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57875232223, "tokens_per_second": 146992.06443223663, "data_wait_s": 1.1618998541962355e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 395500, "loss": 0.6386530287563801, "lr": 0.0003, "grad_norm": 0.15998983031841527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57889856109, "tokens_per_second": 148535.13742564007, "data_wait_s": 9.37400182010606e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 395600, "loss": 0.5625989250838757, "lr": 0.0003, "grad_norm": 0.17024057662099099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57904399230, "tokens_per_second": 147798.79953583484, "data_wait_s": 1.2132004485465586e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3bbf2c3d3f3a2b9d84161597f41486dfe709a1f251d3871949ff3a3511c3d4ab
-size 864219
+oid sha256:0cf4f54a998249f22d36f1ac964a9f73645c038c08239007c9068d2d4b056d4a
+size 866337