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": 76036.33865079588,
-      "required_pages_per_second_upper_estimate": 0.005260370336474848
+      "required_token_positions_per_second": 75894.0307823442,
+      "required_pages_per_second_upper_estimate": 0.005250525147935613
     },
     {
       "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": 9504.542331349485,
-      "required_pages_per_second_upper_estimate": 0.003979353396722531
+      "required_token_positions_per_second": 9486.753847793025,
+      "required_pages_per_second_upper_estimate": 0.003971905730123219
     },
     {
       "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": 19009.08466269897,
-      "required_pages_per_second_upper_estimate": 0.003959091031966455
+      "required_token_positions_per_second": 18973.50769558605,
+      "required_pages_per_second_upper_estimate": 0.00395168128794455
     },
     {
       "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": 28513.626994048453,
-      "required_pages_per_second_upper_estimate": 0.0017093484005057044
+      "required_token_positions_per_second": 28460.261543379078,
+      "required_pages_per_second_upper_estimate": 0.0017061492232224005
     },
     {
       "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": 28513.626994048453,
-      "required_pages_per_second_upper_estimate": 0.0022168090805967234
+      "required_token_positions_per_second": 28460.261543379078,
+      "required_pages_per_second_upper_estimate": 0.00221266015153699
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28513.626994048453,
-      "required_pages_per_second_upper_estimate": 0.36574219153228477
+      "required_token_positions_per_second": 28460.261543379078,
+      "required_pages_per_second_upper_estimate": 0.3650576768304547
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.3159707569609367,
-    "producer_p95_seconds": 0.20482549900043523,
-    "gpu_update_seconds": 0.9696416385670917,
-    "producer_batches_per_second": 3.1648498412263812,
-    "gpu_batches_per_second": 1.031308846717609,
-    "producer_headroom": 3.0687701858655485,
+    "producer_batch_seconds": 0.22949544815629963,
+    "producer_p95_seconds": 0.22115436600142857,
+    "gpu_update_seconds": 0.9714598004610383,
+    "producer_batches_per_second": 4.357384898191716,
+    "gpu_batches_per_second": 1.0293786727205974,
+    "producer_headroom": 4.233024263729266,
     "sustainable": true,
-    "buffer_seconds": 60.11778159115969,
+    "buffer_seconds": 60.23050762858438,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 475510.09989054175,
-    "upload_bytes_per_second": 4764394.786287227,
+    "generation_bytes_per_second": 462487.2676364347,
+    "upload_bytes_per_second": 4706797.530965227,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.03339812791052,
+    "estimated_full_shards_per_hour": 99.23900148219853,
     "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": [
@@ -526,6 +526,12 @@
       "bytes": 6004944,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 3.7895365029999084,
+      "bytes": 13350183,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -110,3 +110,4 @@
 {"step": 731000, "loss": 0.5641935467720032, "lr": 5e-05, "grad_norm": 0.2287241484936114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106997690229, "tokens_per_second": 151364.96428365033, "data_wait_s": 1.3975999536341988e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 731100, "loss": 0.7164812088012695, "lr": 5e-05, "grad_norm": 0.32625489069749375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107012288288, "tokens_per_second": 148174.05047318505, "data_wait_s": 1.487400004407391e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 731200, "loss": 0.5573129653930664, "lr": 5e-05, "grad_norm": 0.22699834790070478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107026740621, "tokens_per_second": 149328.75677081733, "data_wait_s": 1.4477000149781816e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 731300, "loss": 0.5672628283500671, "lr": 5e-05, "grad_norm": 0.22761026205481524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107041254612, "tokens_per_second": 150343.770347667, "data_wait_s": 1.3818998922943138e-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:d3ac4ca648bdf0e2172d5991661829387c0899e9e0d3fd2babe80d3aa41e24ae
-size 238947
+oid sha256:14966ccffeccaf633eb21337d91fb51a6f99280bef1b02ea8a61ad787f218458
+size 241065