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": 75291.58114657442,
-      "required_pages_per_second_upper_estimate": 0.0065260834290172594
+      "required_token_positions_per_second": 75001.19056381253,
+      "required_pages_per_second_upper_estimate": 0.006500913109291662
     },
     {
       "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": 9411.447643321802,
-      "required_pages_per_second_upper_estimate": 0.003754367178602921
+      "required_token_positions_per_second": 9375.148820476566,
+      "required_pages_per_second_upper_estimate": 0.003739887035454191
     },
     {
       "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": 18822.895286643605,
-      "required_pages_per_second_upper_estimate": 0.004354345552922284
+      "required_token_positions_per_second": 18750.297640953133,
+      "required_pages_per_second_upper_estimate": 0.00433755136526672
     },
     {
       "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": 28234.342929965405,
-      "required_pages_per_second_upper_estimate": 0.0017193233502859498
+      "required_token_positions_per_second": 28125.4464614297,
+      "required_pages_per_second_upper_estimate": 0.0017126921266877406
     },
     {
       "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": 28234.342929965405,
-      "required_pages_per_second_upper_estimate": 0.00217355787641228
+      "required_token_positions_per_second": 28125.4464614297,
+      "required_pages_per_second_upper_estimate": 0.002165174724819683
     },
     {
       "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": 28234.342929965405,
-      "required_pages_per_second_upper_estimate": 0.3505760449230218
+      "required_token_positions_per_second": 28125.4464614297,
+      "required_pages_per_second_upper_estimate": 0.34922391523684393
     }
   ],
   "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": 0.2933744920000336,
-    "producer_p95_seconds": 0.20507368499966105,
-    "gpu_update_seconds": 0.9792329883001062,
-    "producer_batches_per_second": 3.408612634256851,
-    "gpu_batches_per_second": 1.0212074265757163,
-    "producer_headroom": 3.337825935800833,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.71244527460659,
+    "buffer_seconds": 60.947512507961925,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366756.0792183762,
-    "upload_bytes_per_second": 4791797.80625389,
+    "generation_bytes_per_second": 385520.2429325754,
+    "upload_bytes_per_second": 5034555.945177843,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.69731695569482,
+    "estimated_full_shards_per_hour": 82.72366968138644,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 1124000,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.3055697850004435,
+      "bytes": 13351938,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -55,3 +55,4 @@
 {"step": 160500, "loss": 0.6106337830424309, "lr": 0.0003, "grad_norm": 0.1615878317458639, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23490310647, "tokens_per_second": 149488.45663925834, "data_wait_s": 1.5161000192165375e-05, "gpu_peak_gb": 85.46427917480469}
 {"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}

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:a18e7f7012cc5963e20fc57a876552eac1aa8aec3af323b0e1605f69dbd6e880
-size 121674
+oid sha256:85fe766cce2d368d9b7b6bd46ea04cefccac0c6ed203b7fa9baa799ed35b0972
+size 123792