Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76196.05664714669,
-      "required_pages_per_second_upper_estimate": 0.005271419998059207
+      "required_token_positions_per_second": 76270.90467646283,
+      "required_pages_per_second_upper_estimate": 0.005276598158398653
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9524.507080893336,
-      "required_pages_per_second_upper_estimate": 0.003987712220445163
+      "required_token_positions_per_second": 9533.863084557854,
+      "required_pages_per_second_upper_estimate": 0.0039916293838039235
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19049.014161786672,
-      "required_pages_per_second_upper_estimate": 0.00396740729361472
+      "required_token_positions_per_second": 19067.726169115707,
+      "required_pages_per_second_upper_estimate": 0.003971304511272541
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28573.52124268001,
-      "required_pages_per_second_upper_estimate": 0.001712938969257944
+      "required_token_positions_per_second": 28601.589253673563,
+      "required_pages_per_second_upper_estimate": 0.0017146216036596363
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28573.52124268001,
-      "required_pages_per_second_upper_estimate": 0.002221465594980869
+      "required_token_positions_per_second": 28601.589253673563,
+      "required_pages_per_second_upper_estimate": 0.002223647759377485
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28573.52124268001,
-      "required_pages_per_second_upper_estimate": 0.36651045064429666
+      "required_token_positions_per_second": 28601.589253673563,
+      "required_pages_per_second_upper_estimate": 0.36687047695223973
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.47277265780459743,
+    "producer_batch_seconds": 0.07810531354695627,
     "producer_p95_seconds": 0.26530443799856585,
-    "gpu_update_seconds": 0.9676091289267643,
-    "producer_batches_per_second": 2.1151815433736694,
-    "gpu_batches_per_second": 1.033475160687211,
-    "producer_headroom": 2.046668970705765,
+    "gpu_update_seconds": 0.9666595710743212,
+    "producer_batches_per_second": 12.803226241436292,
+    "gpu_batches_per_second": 1.0344903520570587,
+    "producer_headroom": 12.3763611869143,
     "sustainable": true,
-    "buffer_seconds": 60.95937512238615,
+    "buffer_seconds": 60.899552977682234,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 255132.95679941511,
-    "upload_bytes_per_second": 5321122.045337513,
+    "generation_bytes_per_second": 254182.0198320507,
+    "upload_bytes_per_second": 5248181.285811385,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.74559333788719,
+    "estimated_full_shards_per_hour": 54.54154440137044,
     "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": 1.2612079769969569,
-      "bytes": 16881065,
-      "pages": 36,
-      "shards": 5
-    },
     {
       "seconds": 1.1940166240019607,
       "bytes": 6068573,
@@ -562,6 +556,12 @@
       "bytes": 812811,
       "pages": 23,
       "shards": 1
+    },
+    {
+      "seconds": 1.1825939109985484,
+      "bytes": 10557366,
+      "pages": 46,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -294,3 +294,4 @@
 {"step": 749400, "loss": 0.497047483921051, "lr": 5e-05, "grad_norm": 0.2345468090923336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109691158136, "tokens_per_second": 149560.74796764064, "data_wait_s": 8.914998034015298e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 749500, "loss": 0.5131856799125671, "lr": 5e-05, "grad_norm": 0.2493631220042696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109705650624, "tokens_per_second": 146846.81138012517, "data_wait_s": 1.6113001038320363e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 749600, "loss": 0.5182501673698425, "lr": 5e-05, "grad_norm": 0.2333670310636725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109720128723, "tokens_per_second": 149135.69141196576, "data_wait_s": 1.3901000784244388e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 749700, "loss": 0.5342912077903748, "lr": 5e-05, "grad_norm": 0.2368940682671192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109734841210, "tokens_per_second": 151074.9557897747, "data_wait_s": 1.3933000445831567e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ade489bafb695cd58fa3da4ee2f65f466e4ead0bfa566d7550a49c085a07d057
-size 631008
+oid sha256:949551a8fb354801ccfb585d6eb4e4d9c2a01ff922c0d20a34c74e0b8fb4c344
+size 633126