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": 75709.36135918561,
-      "required_pages_per_second_upper_estimate": 0.00656229555899213
+      "required_token_positions_per_second": 76126.24436217148,
+      "required_pages_per_second_upper_estimate": 0.0065984299211106325
     },
     {
       "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": 9463.670169898201,
-      "required_pages_per_second_upper_estimate": 0.003775199525250599
+      "required_token_positions_per_second": 9515.780545271435,
+      "required_pages_per_second_upper_estimate": 0.0037959871331065243
     },
     {
       "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": 18927.340339796403,
-      "required_pages_per_second_upper_estimate": 0.0043785070777990305
+      "required_token_positions_per_second": 19031.56109054287,
+      "required_pages_per_second_upper_estimate": 0.004402616714261658
     },
     {
       "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": 28391.010509694603,
-      "required_pages_per_second_upper_estimate": 0.0017288635839201926
+      "required_token_positions_per_second": 28547.341635814304,
+      "required_pages_per_second_upper_estimate": 0.0017383833292948525
     },
     {
       "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": 28391.010509694603,
-      "required_pages_per_second_upper_estimate": 0.0021856185803834512
+      "required_token_positions_per_second": 28547.341635814304,
+      "required_pages_per_second_upper_estimate": 0.0021976533832244005
     },
     {
       "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": 28391.010509694603,
-      "required_pages_per_second_upper_estimate": 0.35252133193059837
+      "required_token_positions_per_second": 28547.341635814304,
+      "required_pages_per_second_upper_estimate": 0.35446244131038285
     }
   ],
   "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": 0.3390095536171671,
-    "producer_p95_seconds": 0.19312582800012024,
-    "gpu_update_seconds": 0.9738293742858892,
-    "producer_batches_per_second": 2.9497693776773874,
-    "gpu_batches_per_second": 1.0268739333656902,
-    "producer_headroom": 2.872572067351247,
+    "producer_batch_seconds": 0.07357052367189709,
+    "producer_p95_seconds": 0.174305973000628,
+    "gpu_update_seconds": 0.9684964839357922,
+    "producer_batches_per_second": 13.592400190865924,
+    "gpu_batches_per_second": 1.0325282709712928,
+    "producer_headroom": 13.164191793101839,
     "sustainable": true,
-    "buffer_seconds": 60.377421205725135,
+    "buffer_seconds": 60.04678200401912,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382617.8944388639,
-    "upload_bytes_per_second": 4686271.013316097,
+    "generation_bytes_per_second": 382704.3369469629,
+    "upload_bytes_per_second": 4931732.628185666,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.10089325785101,
+    "estimated_full_shards_per_hour": 82.11944180781046,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 968274,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 1.2902597590000369,
+      "bytes": 12650540,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -52,3 +52,4 @@
 {"step": 160200, "loss": 0.6043591499328613, "lr": 0.0003, "grad_norm": 0.14702501515498978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23446262115, "tokens_per_second": 149684.11786736164, "data_wait_s": 1.3070999557385221e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 160300, "loss": 0.6843377538025379, "lr": 0.0003, "grad_norm": 0.1873446259741128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23460885944, "tokens_per_second": 149292.79984657269, "data_wait_s": 1.415899987478042e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 160400, "loss": 0.5886963568627834, "lr": 0.0003, "grad_norm": 0.1746030324599597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23475621908, "tokens_per_second": 151076.70153386504, "data_wait_s": 1.3455000043904874e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 160500, "loss": 0.6106337830424309, "lr": 0.0003, "grad_norm": 0.1615878317458639, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23490310647, "tokens_per_second": 149488.45663925834, "data_wait_s": 1.5161000192165375e-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 sha256:13cbb773c886045f7bb99ac5f396ea0a03c6b32fa6c1ae97014e5be191979646
-size 115320
+oid sha256:a2410d01dd4f8c4318d9e9f90625e2a34f3fbba751a9870248e977db16294574
+size 117438