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": 77242.14292976781,
-      "required_pages_per_second_upper_estimate": 0.005343790674345522
+      "required_token_positions_per_second": 76676.52509121057,
+      "required_pages_per_second_upper_estimate": 0.005304659919859931
     },
     {
       "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": 9655.267866220976,
-      "required_pages_per_second_upper_estimate": 0.004042459030666142
+      "required_token_positions_per_second": 9584.565636401321,
+      "required_pages_per_second_upper_estimate": 0.004012857483471102
     },
     {
       "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": 19310.535732441953,
-      "required_pages_per_second_upper_estimate": 0.004021875339994608
+      "required_token_positions_per_second": 19169.131272802642,
+      "required_pages_per_second_upper_estimate": 0.003992424520138111
     },
     {
       "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": 28965.80359866293,
-      "required_pages_per_second_upper_estimate": 0.0017364556975186444
+      "required_token_positions_per_second": 28753.69690920396,
+      "required_pages_per_second_upper_estimate": 0.0017237402253537416
     },
     {
       "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": 28965.80359866293,
-      "required_pages_per_second_upper_estimate": 0.00225196382269081
+      "required_token_positions_per_second": 28753.69690920396,
+      "required_pages_per_second_upper_estimate": 0.002235473460544104
     },
     {
       "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": 28965.80359866293,
-      "required_pages_per_second_upper_estimate": 0.37154222750686794
+      "required_token_positions_per_second": 28753.69690920396,
+      "required_pages_per_second_upper_estimate": 0.3688215506369077
     }
   ],
   "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": 0.13528335564086547,
-    "producer_p95_seconds": 0.3221259400015697,
-    "gpu_update_seconds": 0.9545048493415954,
-    "producer_batches_per_second": 7.391892337847413,
-    "gpu_batches_per_second": 1.0476636139562692,
-    "producer_headroom": 7.055597082286338,
+    "producer_batch_seconds": 0.3875471729609501,
+    "producer_p95_seconds": 0.32004022000182886,
+    "gpu_update_seconds": 0.9615459218097958,
+    "producer_batches_per_second": 2.5803310403731463,
+    "gpu_batches_per_second": 1.0399919310331294,
+    "producer_headroom": 2.481106788790026,
     "sustainable": true,
-    "buffer_seconds": 60.13380550852051,
+    "buffer_seconds": 60.57739307401714,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320423.1562543524,
-    "upload_bytes_per_second": 5271609.199435923,
+    "generation_bytes_per_second": 314519.1307451434,
+    "upload_bytes_per_second": 5139842.369727401,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.75535026286057,
+    "estimated_full_shards_per_hour": 67.48848382726409,
     "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": 1.1806990879995283,
-      "bytes": 12168625,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 0.9337092679998023,
       "bytes": 919576,
@@ -562,6 +556,12 @@
       "bytes": 7658131,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.01357181700223,
+      "bytes": 644212,
+      "pages": 18,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -326,3 +326,4 @@
 {"step": 637000, "loss": 0.5176790952682495, "lr": 0.0001, "grad_norm": 0.17257702674561864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93240168892, "tokens_per_second": 152955.16928854142, "data_wait_s": 1.0569998266873881e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 637100, "loss": 0.5054295659065247, "lr": 0.0001, "grad_norm": 0.21669725697026537, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93254994326, "tokens_per_second": 153752.79937595213, "data_wait_s": 1.6162997781066224e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 637200, "loss": 0.56538325548172, "lr": 0.0001, "grad_norm": 0.16798120802493155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93269558678, "tokens_per_second": 151645.69934552547, "data_wait_s": 1.0137999197468162e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 637300, "loss": 0.556537926197052, "lr": 0.0001, "grad_norm": 0.1721248058661298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93284114297, "tokens_per_second": 147904.51063344016, "data_wait_s": 8.86399720911868e-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:df58623695aeee3e7f3b167554084e271c89c211ead4210c0bf3e093db794945
-size 700350
+oid sha256:060e3fcbb8a4c9359535c1840b999da514b413fd7be2b1de65e39610ecb5db84
+size 702468