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": 74683.6994578,
-      "required_pages_per_second_upper_estimate": 0.006473393784896348
+      "required_token_positions_per_second": 74699.91833092092,
+      "required_pages_per_second_upper_estimate": 0.0064747995957121105
     },
     {
       "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": 9335.462432225,
-      "required_pages_per_second_upper_estimate": 0.00372405554181626
+      "required_token_positions_per_second": 9337.489791365115,
+      "required_pages_per_second_upper_estimate": 0.003724864285688972
     },
     {
       "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": 18670.92486445,
-      "required_pages_per_second_upper_estimate": 0.0043191898703358765
+      "required_token_positions_per_second": 18674.97958273023,
+      "required_pages_per_second_upper_estimate": 0.004320127858049402
     },
     {
       "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": 28006.387296675002,
-      "required_pages_per_second_upper_estimate": 0.0017054420482093412
+      "required_token_positions_per_second": 28012.469374095348,
+      "required_pages_per_second_upper_estimate": 0.001705812414814046
     },
     {
       "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": 28006.387296675002,
-      "required_pages_per_second_upper_estimate": 0.002156009220740005
+      "required_token_positions_per_second": 28012.469374095348,
+      "required_pages_per_second_upper_estimate": 0.002156477435896091
     },
     {
       "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": 28006.387296675002,
-      "required_pages_per_second_upper_estimate": 0.3477455988759825
+      "required_token_positions_per_second": 28012.469374095348,
+      "required_pages_per_second_upper_estimate": 0.3478211179221395
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11815777263281291,
-    "producer_p95_seconds": 0.23221001899946714,
-    "gpu_update_seconds": 0.9872033728278281,
-    "producer_batches_per_second": 8.463260416287635,
-    "gpu_batches_per_second": 1.0129625034966363,
-    "producer_headroom": 8.354959228079403,
+    "producer_batch_seconds": 0.05888063563281776,
+    "producer_p95_seconds": 0.20180956800140848,
+    "gpu_update_seconds": 0.9869890308766427,
+    "producer_batches_per_second": 16.98351230846155,
+    "gpu_batches_per_second": 1.0131824860422218,
+    "producer_headroom": 16.762540354209996,
     "sustainable": true,
-    "buffer_seconds": 60.219405742497514,
+    "buffer_seconds": 60.206330883475204,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349332.024435776,
-    "upload_bytes_per_second": 5294312.580484009,
+    "generation_bytes_per_second": 349718.9336879294,
+    "upload_bytes_per_second": 5260384.973756027,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.95852041058502,
+    "estimated_full_shards_per_hour": 75.04154212931073,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 7640268,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.723643687000731,
+      "bytes": 1209843,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -103,3 +103,4 @@
 {"step": 205300, "loss": 0.584257559850812, "lr": 0.0003, "grad_norm": 0.14663389437398983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30048220354, "tokens_per_second": 148249.03032386256, "data_wait_s": 1.8435999663779512e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 205400, "loss": 0.6149740368127823, "lr": 0.0003, "grad_norm": 0.13280079023353883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30062883332, "tokens_per_second": 147862.43994892322, "data_wait_s": 2.2259999241214246e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 205500, "loss": 0.6220797058194876, "lr": 0.0003, "grad_norm": 0.14416894672017122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30077606188, "tokens_per_second": 148590.61643834022, "data_wait_s": 1.797600089048501e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 205600, "loss": 0.5864623375236988, "lr": 0.0003, "grad_norm": 0.17855335037674847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30092330583, "tokens_per_second": 148678.9822647812, "data_wait_s": 1.805900137696881e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12be4698baa844339244e2a435a21e8c58f67007103ed61d0e79f4744594e315
-size 224121
+oid sha256:163321637941f018e7c6c7d68b170a099aaec7880795c299616a68a653fd5491
+size 226239