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": 76149.08912370521,
-      "required_pages_per_second_upper_estimate": 0.006600410047140858
+      "required_token_positions_per_second": 75070.03584646276,
+      "required_pages_per_second_upper_estimate": 0.006506880443905011
     },
     {
       "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": 9518.636140463152,
-      "required_pages_per_second_upper_estimate": 0.003797126272723453
+      "required_token_positions_per_second": 9383.754480807846,
+      "required_pages_per_second_upper_estimate": 0.003743319962026426
     },
     {
       "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": 19037.272280926303,
-      "required_pages_per_second_upper_estimate": 0.004403937897643348
+      "required_token_positions_per_second": 18767.50896161569,
+      "required_pages_per_second_upper_estimate": 0.004341532901393105
     },
     {
       "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": 28555.908421389457,
-      "required_pages_per_second_upper_estimate": 0.001738905001589929
+      "required_token_positions_per_second": 28151.263442423537,
+      "required_pages_per_second_upper_estimate": 0.0017142642453790342
     },
     {
       "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": 28555.908421389457,
-      "required_pages_per_second_upper_estimate": 0.0021983128780923558
+      "required_token_positions_per_second": 28151.263442423537,
+      "required_pages_per_second_upper_estimate": 0.002167162187482566
     },
     {
       "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": 28555.908421389457,
-      "required_pages_per_second_upper_estimate": 0.3545688121160393
+      "required_token_positions_per_second": 28151.263442423537,
+      "required_pages_per_second_upper_estimate": 0.3495444757369102
     }
   ],
   "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": 0.12990309982814097,
-    "producer_p95_seconds": 0.396793042003992,
-    "gpu_update_seconds": 0.9682059345480531,
-    "producer_batches_per_second": 7.698045707323218,
-    "gpu_batches_per_second": 1.0328381228801164,
-    "producer_headroom": 7.453293538252505,
+    "producer_batch_seconds": 0.26545161323451794,
+    "producer_p95_seconds": 0.3775601310044294,
+    "gpu_update_seconds": 0.9821228825678522,
+    "producer_batches_per_second": 3.7671648999041203,
+    "gpu_batches_per_second": 1.0182025261293235,
+    "producer_headroom": 3.699818850602269,
     "sustainable": true,
-    "buffer_seconds": 60.0287679419793,
+    "buffer_seconds": 60.89161871920683,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 175,
+    "upload_pages_per_commit": 158,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 582583.5312291479,
+    "generation_bytes_per_second": 577337.1695354284,
     "upload_bytes_per_second": 4350860.232751541,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.504431975630894,
+    "estimated_full_shards_per_hour": 61.94155843041963,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -414,3 +414,4 @@
 {"step": 196400, "loss": 0.6060383096337318, "lr": 0.0003, "grad_norm": 0.15123548976325515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28744944508, "tokens_per_second": 150985.67238656757, "data_wait_s": 1.1547999747563154e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196500, "loss": 0.5728410333395004, "lr": 0.0003, "grad_norm": 0.13311416429812908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28759668846, "tokens_per_second": 150816.59064497318, "data_wait_s": 8.973998774308711e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 196600, "loss": 0.5985371414572, "lr": 0.0003, "grad_norm": 0.14992750001195332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28774382558, "tokens_per_second": 150662.8953278723, "data_wait_s": 1.0135001502931118e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 196700, "loss": 0.6548292003571987, "lr": 0.0003, "grad_norm": 0.16110462392309002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28789175919, "tokens_per_second": 151406.54491275083, "data_wait_s": 1.012500433716923e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f66874b99fc11964964c3fb3e0f913c90331eb9eb640b96f97cc96a65a418ea
-size 887517
+oid sha256:df302ae3a17ee69d3a9fc45ba093417896dde4db57468b82603349e2fbf113d8
+size 889635