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": 75001.19056381253,
-      "required_pages_per_second_upper_estimate": 0.006500913109291662
+      "required_token_positions_per_second": 76070.88230945659,
+      "required_pages_per_second_upper_estimate": 0.006593631278695146
     },
     {
       "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": 9375.148820476566,
-      "required_pages_per_second_upper_estimate": 0.003739887035454191
+      "required_token_positions_per_second": 9508.860288682074,
+      "required_pages_per_second_upper_estimate": 0.0037932265392859716
     },
     {
       "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": 18750.297640953133,
-      "required_pages_per_second_upper_estimate": 0.00433755136526672
+      "required_token_positions_per_second": 19017.720577364147,
+      "required_pages_per_second_upper_estimate": 0.004399414955122473
     },
     {
       "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": 28125.4464614297,
-      "required_pages_per_second_upper_estimate": 0.0017126921266877406
+      "required_token_positions_per_second": 28526.580866046217,
+      "required_pages_per_second_upper_estimate": 0.0017371191073393168
     },
     {
       "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": 28125.4464614297,
-      "required_pages_per_second_upper_estimate": 0.002165174724819683
+      "required_token_positions_per_second": 28526.580866046217,
+      "required_pages_per_second_upper_estimate": 0.002196055161698164
     },
     {
       "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": 28125.4464614297,
-      "required_pages_per_second_upper_estimate": 0.34922391523684393
+      "required_token_positions_per_second": 28526.580866046217,
+      "required_pages_per_second_upper_estimate": 0.35420466203169
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06727939848445175,
-    "producer_p95_seconds": 0.1770332390005933,
-    "gpu_update_seconds": 0.9830243952897084,
-    "producer_batches_per_second": 14.863390912020412,
-    "gpu_batches_per_second": 1.0172687522218498,
-    "producer_headroom": 14.611075863243414,
+    "producer_batch_seconds": 0.07022832300002335,
+    "producer_p95_seconds": 0.18761593799990806,
+    "gpu_update_seconds": 0.9692013259432731,
+    "producer_batches_per_second": 14.23926924753233,
+    "gpu_batches_per_second": 1.0317773750740096,
+    "producer_headroom": 13.800718635171608,
     "sustainable": true,
-    "buffer_seconds": 60.947512507961925,
+    "buffer_seconds": 60.09048220848293,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385520.2429325754,
-    "upload_bytes_per_second": 5034555.945177843,
+    "generation_bytes_per_second": 385922.01699967333,
+    "upload_bytes_per_second": 5067017.63386749,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.72366968138644,
+    "estimated_full_shards_per_hour": 82.80988104336404,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 13351938,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.0015427250000357,
+      "bytes": 6023702,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -56,3 +56,4 @@
 {"step": 160600, "loss": 0.6279516518115997, "lr": 0.0003, "grad_norm": 0.13974949827649824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23505056242, "tokens_per_second": 150770.0126285993, "data_wait_s": 1.2548999620776158e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 160700, "loss": 0.6313633881509304, "lr": 0.0003, "grad_norm": 0.14442777504246337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23519771939, "tokens_per_second": 149875.528235883, "data_wait_s": 2.9269999686221126e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 160800, "loss": 0.5601533055305481, "lr": 0.0003, "grad_norm": 0.15310807497503692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23534491326, "tokens_per_second": 150268.9764084363, "data_wait_s": 9.97200004348997e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 160900, "loss": 0.604211675003171, "lr": 0.0003, "grad_norm": 0.14165015113283677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23548984978, "tokens_per_second": 148113.15121979677, "data_wait_s": 9.25599943002453e-06, "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 sha256:85fe766cce2d368d9b7b6bd46ea04cefccac0c6ed203b7fa9baa799ed35b0972
-size 123792
+oid sha256:28d31db0447d7fea1ff52d03f64fe5c28f4cb5db3f8b504b42cbe10271846c62
+size 125910