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": 77284.53832720044,
-      "required_pages_per_second_upper_estimate": 0.005346723686310786
+      "required_token_positions_per_second": 77112.01444718293,
+      "required_pages_per_second_upper_estimate": 0.005334788083980621
     },
     {
       "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": 9660.567290900055,
-      "required_pages_per_second_upper_estimate": 0.004044677789114701
+      "required_token_positions_per_second": 9639.001805897866,
+      "required_pages_per_second_upper_estimate": 0.0040356487708828204
     },
     {
       "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": 19321.13458180011,
-      "required_pages_per_second_upper_estimate": 0.004024082800805458
+      "required_token_positions_per_second": 19278.00361179573,
+      "required_pages_per_second_upper_estimate": 0.004015099757193715
     },
     {
       "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": 28981.701872700163,
-      "required_pages_per_second_upper_estimate": 0.0017374087747719178
+      "required_token_positions_per_second": 28917.0054176936,
+      "required_pages_per_second_upper_estimate": 0.0017335303210800413
     },
     {
       "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": 28981.701872700163,
-      "required_pages_per_second_upper_estimate": 0.0022531998435680878
+      "required_token_positions_per_second": 28917.0054176936,
+      "required_pages_per_second_upper_estimate": 0.002248169978760959
     },
     {
       "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": 28981.701872700163,
-      "required_pages_per_second_upper_estimate": 0.37174615349598084
+      "required_token_positions_per_second": 28917.0054176936,
+      "required_pages_per_second_upper_estimate": 0.3709162968367979
     }
   ],
   "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": 0.05311000085946205,
-    "producer_p95_seconds": 0.17998047400033101,
-    "gpu_update_seconds": 0.9539812438014046,
-    "producer_batches_per_second": 18.828845487051815,
-    "gpu_batches_per_second": 1.0482386383355093,
-    "producer_headroom": 17.962365437082155,
+    "producer_batch_seconds": 0.4026988276797283,
+    "producer_p95_seconds": 0.2175319979996857,
+    "gpu_update_seconds": 0.9561156005138373,
+    "producer_batches_per_second": 2.483245371638661,
+    "gpu_batches_per_second": 1.0458986334524594,
+    "producer_headroom": 2.374269639727505,
     "sustainable": true,
-    "buffer_seconds": 60.100818359488485,
+    "buffer_seconds": 60.235282832371745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373287.5368180717,
-    "upload_bytes_per_second": 5603932.082219664,
+    "generation_bytes_per_second": 337350.52917814086,
+    "upload_bytes_per_second": 5715018.8906609155,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.09881571203816,
+    "estimated_full_shards_per_hour": 72.3875704432313,
     "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": 0.8495643409999047,
-      "bytes": 825792,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.5516937659999712,
       "bytes": 10821331,
@@ -562,6 +556,12 @@
       "bytes": 1126127,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.2304532270009076,
+      "bytes": 11321849,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -85,3 +85,4 @@
 {"step": 612900, "loss": 0.5587241053581238, "lr": 0.0001, "grad_norm": 0.1471157718397796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89710465434, "tokens_per_second": 151871.62118696576, "data_wait_s": 2.951700116682332e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 613000, "loss": 0.5719670653343201, "lr": 0.0001, "grad_norm": 0.14760701833825796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89725241263, "tokens_per_second": 153968.85745948154, "data_wait_s": 9.84799953585025e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 613100, "loss": 0.5231931209564209, "lr": 0.0001, "grad_norm": 0.14752648244788277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89739912364, "tokens_per_second": 152319.68489424567, "data_wait_s": 1.1221000022487715e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 613200, "loss": 0.5482064485549927, "lr": 0.0001, "grad_norm": 0.14933810984877444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89754711284, "tokens_per_second": 153687.61228747087, "data_wait_s": 9.2249993031146e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:88fc11b22606a8f551bcbd9a642f570c11dae1d51407cfa6fd45318f3712ca51
-size 185997
+oid sha256:8c4704f243edcb343b8bd832998c0284e6b5feb0aae455669244d11ecfa3ec24
+size 188115