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": 73913.29132246837,
-      "required_pages_per_second_upper_estimate": 0.0064066167603073695
+      "required_token_positions_per_second": 73862.86564174516,
+      "required_pages_per_second_upper_estimate": 0.006402245990105001
     },
     {
       "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": 9239.161415308547,
-      "required_pages_per_second_upper_estimate": 0.0036856396263397744
+      "required_token_positions_per_second": 9232.858205218145,
+      "required_pages_per_second_upper_estimate": 0.003683125181593324
     },
     {
       "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": 18478.322830617093,
-      "required_pages_per_second_upper_estimate": 0.004274634779488657
+      "required_token_positions_per_second": 18465.71641043629,
+      "required_pages_per_second_upper_estimate": 0.0042717185060452455
     },
     {
       "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": 27717.48424592564,
-      "required_pages_per_second_upper_estimate": 0.0016878493681758684
+      "required_token_positions_per_second": 27698.574615654434,
+      "required_pages_per_second_upper_estimate": 0.0016866978709035145
     },
     {
       "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": 27717.48424592564,
-      "required_pages_per_second_upper_estimate": 0.002133768664158489
+      "required_token_positions_per_second": 27698.574615654434,
+      "required_pages_per_second_upper_estimate": 0.0021323129484750046
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27717.48424592564,
-      "required_pages_per_second_upper_estimate": 0.34415838988198766
+      "required_token_positions_per_second": 27698.574615654434,
+      "required_pages_per_second_upper_estimate": 0.3439235955604807
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4006766042501795,
-    "producer_p95_seconds": 0.3965510100024403,
-    "gpu_update_seconds": 0.9974931258079148,
-    "producer_batches_per_second": 2.4957783643778897,
-    "gpu_batches_per_second": 1.0025131744041391,
-    "producer_headroom": 2.489521762007066,
+    "producer_batch_seconds": 0.10133812023451583,
+    "producer_p95_seconds": 0.38650077000056626,
+    "gpu_update_seconds": 0.9981741076443027,
+    "producer_batches_per_second": 9.867954898766707,
+    "gpu_batches_per_second": 1.0018292323370384,
+    "producer_headroom": 9.849937075350683,
     "sustainable": true,
-    "buffer_seconds": 60.84708067428281,
+    "buffer_seconds": 60.888620566302464,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348860.6910331774,
+    "generation_bytes_per_second": 347867.2680437472,
     "upload_bytes_per_second": 5817052.546362575,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.85738323446743,
+    "estimated_full_shards_per_hour": 74.64421778663932,
     "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

@@ -370,3 +370,4 @@
 {"step": 232000, "loss": 0.6067023389041424, "lr": 0.0003, "grad_norm": 0.1522415591060405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33958678080, "tokens_per_second": 148290.96934798567, "data_wait_s": 1.2838994734920561e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 232100, "loss": 0.5992082916200161, "lr": 0.0003, "grad_norm": 0.1648074949305529, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33973124088, "tokens_per_second": 146084.34500625602, "data_wait_s": 1.507100387243554e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 232200, "loss": 0.5945616010576487, "lr": 0.0003, "grad_norm": 0.1442751482313228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33987578625, "tokens_per_second": 146378.1621474991, "data_wait_s": 1.733299723127857e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 232300, "loss": 0.6218109801411629, "lr": 0.0003, "grad_norm": 0.1415363212380993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34002106053, "tokens_per_second": 148234.62647555297, "data_wait_s": 1.3174998457543552e-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:ba154a3ac16db371583a4d084cb90b0d46013ae28debc61f82ac86c8a7b278fb
-size 793542
+oid sha256:7c0ff349f1bd95fcba6deb082a1b6d6284ee25314a1e78d8ef41794096a4a13c
+size 795660