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": 74872.19943677516,
-      "required_pages_per_second_upper_estimate": 0.0064897324853250855
+      "required_token_positions_per_second": 74744.0276601196,
+      "required_pages_per_second_upper_estimate": 0.006478622880572979
     },
     {
       "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": 9359.024929596895,
-      "required_pages_per_second_upper_estimate": 0.0037334549743086387
+      "required_token_positions_per_second": 9343.00345751495,
+      "required_pages_per_second_upper_estimate": 0.003727063769552796
     },
     {
       "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": 18718.04985919379,
-      "required_pages_per_second_upper_estimate": 0.004330091408498271
+      "required_token_positions_per_second": 18686.0069150299,
+      "required_pages_per_second_upper_estimate": 0.004322678837302511
     },
     {
       "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": 28077.074788790687,
-      "required_pages_per_second_upper_estimate": 0.0017097465456105776
+      "required_token_positions_per_second": 28029.01037254485,
+      "required_pages_per_second_upper_estimate": 0.001706819674835709
     },
     {
       "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": 28077.074788790687,
-      "required_pages_per_second_upper_estimate": 0.002161450939558572
+      "required_token_positions_per_second": 28029.01037254485,
+      "required_pages_per_second_upper_estimate": 0.0021577508077451476
     },
     {
       "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": 28077.074788790687,
-      "required_pages_per_second_upper_estimate": 0.34862330095224164
+      "required_token_positions_per_second": 28029.01037254485,
+      "required_pages_per_second_upper_estimate": 0.34802650176372163
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06762013767189501,
-    "producer_p95_seconds": 0.20516962799956673,
-    "gpu_update_seconds": 0.9847179668103464,
-    "producer_batches_per_second": 14.78849399645086,
-    "gpu_batches_per_second": 1.015519198089941,
-    "producer_headroom": 14.562495740372105,
+    "producer_batch_seconds": 0.12252453838279109,
+    "producer_p95_seconds": 0.2942225980004878,
+    "gpu_update_seconds": 0.9864065706394665,
+    "producer_batches_per_second": 8.161630422763158,
+    "gpu_batches_per_second": 1.013780757108827,
+    "producer_headroom": 8.050685876144545,
     "sustainable": true,
-    "buffer_seconds": 60.06779597543113,
+    "buffer_seconds": 60.17080080900745,
     "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": 428682.5490002683,
-    "upload_bytes_per_second": 5225926.143782697,
+    "generation_bytes_per_second": 430583.4642981231,
+    "upload_bytes_per_second": 5174292.444791185,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.98529579645191,
+    "estimated_full_shards_per_hour": 92.39318796832818,
     "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": 1.0256425439993109,
-      "bytes": 848800,
-      "pages": 26,
-      "shards": 1
-    },
-    {
-      "seconds": 1.479707024000163,
-      "bytes": 13318632,
-      "pages": 43,
-      "shards": 4
-    },
     {
       "seconds": 1.157918292999966,
       "bytes": 10292312,
@@ -562,6 +550,18 @@
       "bytes": 7611150,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0686068680006429,
+      "bytes": 1192521,
+      "pages": 33,
+      "shards": 1
+    },
+    {
+      "seconds": 1.109725468999386,
+      "bytes": 7035828,
+      "pages": 21,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 169800, "loss": 0.5832123067229986, "lr": 0.0003, "grad_norm": 0.14845787700264634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24850943091, "tokens_per_second": 150583.05351849974, "data_wait_s": 1.3396000213106163e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 169900, "loss": 0.5867737848311663, "lr": 0.0003, "grad_norm": 0.13999979719676398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24865609258, "tokens_per_second": 149161.79099476011, "data_wait_s": 1.3983999451738782e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 170000, "loss": 0.5951601397246122, "lr": 0.0003, "grad_norm": 0.13240837884355422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24880314526, "tokens_per_second": 150478.4375103592, "data_wait_s": 1.1244999768678099e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 170100, "loss": 0.6039816215634346, "lr": 0.0003, "grad_norm": 0.13955668703862745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24895107560, "tokens_per_second": 135326.1570778702, "data_wait_s": 1.3150000086170621e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16ed46b37e985f6d2c8ee1a0a56b50edad3ac8224341b9e7a60ea79b765e9937
-size 320214
+oid sha256:13f4f50328f53f0e4327e76490018168038696b67555f7c93f944f8a087594b5
+size 322332