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": 75460.03911019991,
-      "required_pages_per_second_upper_estimate": 0.006540684938351512
+      "required_token_positions_per_second": 75310.7617309654,
+      "required_pages_per_second_upper_estimate": 0.006527745953459523
     },
     {
       "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": 9432.504888774989,
-      "required_pages_per_second_upper_estimate": 0.003762767228648073
+      "required_token_positions_per_second": 9413.845216370675,
+      "required_pages_per_second_upper_estimate": 0.003755323606339028
     },
     {
       "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": 18865.009777549978,
-      "required_pages_per_second_upper_estimate": 0.004364087999203748
+      "required_token_positions_per_second": 18827.69043274135,
+      "required_pages_per_second_upper_estimate": 0.004355454825580309
     },
     {
       "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": 28297.51466632497,
-      "required_pages_per_second_upper_estimate": 0.0017231701775937613
+      "required_token_positions_per_second": 28241.535649112026,
+      "required_pages_per_second_upper_estimate": 0.0017197613491446982
     },
     {
       "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": 28297.51466632497,
-      "required_pages_per_second_upper_estimate": 0.0021784210115477977
+      "required_token_positions_per_second": 28241.535649112026,
+      "required_pages_per_second_upper_estimate": 0.002174111591842908
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28297.51466632497,
-      "required_pages_per_second_upper_estimate": 0.35136042646640636
+      "required_token_positions_per_second": 28241.535649112026,
+      "required_pages_per_second_upper_estimate": 0.35066535442234037
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10195648138287083,
-    "producer_p95_seconds": 0.331178528998862,
-    "gpu_update_seconds": 0.9770469359594356,
-    "producer_batches_per_second": 9.808106227643952,
-    "gpu_batches_per_second": 1.0234922839382583,
-    "producer_headroom": 9.582980137284181,
+    "producer_batch_seconds": 0.43096066231228747,
+    "producer_p95_seconds": 0.3448503260005964,
+    "gpu_update_seconds": 0.9789835915268054,
+    "producer_batches_per_second": 2.3203973992303015,
+    "gpu_batches_per_second": 1.021467579901332,
+    "producer_headroom": 2.271630979667939,
     "sustainable": true,
-    "buffer_seconds": 60.576910029485006,
+    "buffer_seconds": 60.69698267466193,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344005.29731203354,
+    "generation_bytes_per_second": 297306.4717730784,
     "upload_bytes_per_second": 2793934.5555964806,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.81552876969103,
+    "estimated_full_shards_per_hour": 63.79504790205253,
     "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": [

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

@@ -276,3 +276,4 @@
 {"step": 222600, "loss": 0.5505469087511301, "lr": 0.0003, "grad_norm": 0.13772615889590778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32581643509, "tokens_per_second": 148935.62317157237, "data_wait_s": 1.1741001799236983e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 222700, "loss": 0.6319021750241518, "lr": 0.0003, "grad_norm": 0.16434841942010975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32596219237, "tokens_per_second": 147673.46251698988, "data_wait_s": 1.1268999514868483e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 222800, "loss": 0.5395330935716629, "lr": 0.0003, "grad_norm": 0.15129146284377312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32611032732, "tokens_per_second": 149619.1039673062, "data_wait_s": 1.046399847837165e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 222900, "loss": 0.5923827104270458, "lr": 0.0003, "grad_norm": 0.14402803898302785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32625667614, "tokens_per_second": 149895.4228098872, "data_wait_s": 1.2005999451503158e-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:c0167932c656e4e4fbab45a0979328c0f2ea279e7c567d5c222625111fd351b1
-size 592884
+oid sha256:9e367c287ad22eedbe045dddc071bf9965183ddaf006d4a0b580508cad4c06f3
+size 595002