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": 75751.04195723658,
-      "required_pages_per_second_upper_estimate": 0.0065659083275926775
+      "required_token_positions_per_second": 75599.14960961684,
+      "required_pages_per_second_upper_estimate": 0.006552742683868634
     },
     {
       "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": 9468.880244654572,
-      "required_pages_per_second_upper_estimate": 0.0037772779019684746
+      "required_token_positions_per_second": 9449.893701202105,
+      "required_pages_per_second_upper_estimate": 0.0037697038859111636
     },
     {
       "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": 18937.760489309145,
-      "required_pages_per_second_upper_estimate": 0.004380917596000435
+      "required_token_positions_per_second": 18899.78740240421,
+      "required_pages_per_second_upper_estimate": 0.0043721331906484825
     },
     {
       "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": 28406.64073396372,
-      "required_pages_per_second_upper_estimate": 0.0017298153825727326
+      "required_token_positions_per_second": 28349.68110360631,
+      "required_pages_per_second_upper_estimate": 0.001726346839928051
     },
     {
       "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": 28406.64073396372,
-      "required_pages_per_second_upper_estimate": 0.0021868218383149177
+      "required_token_positions_per_second": 28349.68110360631,
+      "required_pages_per_second_upper_estimate": 0.0021824369282956563
     },
     {
       "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": 28406.64073396372,
-      "required_pages_per_second_upper_estimate": 0.35271540700502524
+      "required_token_positions_per_second": 28349.68110360631,
+      "required_pages_per_second_upper_estimate": 0.35200815902760607
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13039045733597732,
-    "producer_p95_seconds": 0.40213196400145534,
-    "gpu_update_seconds": 0.9732935428349798,
-    "producer_batches_per_second": 7.669272893362881,
-    "gpu_batches_per_second": 1.0274392626578313,
-    "producer_headroom": 7.464453785349434,
+    "producer_batch_seconds": 0.12221520023439325,
+    "producer_p95_seconds": 0.4268946180018247,
+    "gpu_update_seconds": 0.9752490653760104,
+    "producer_batches_per_second": 8.182288275780156,
+    "gpu_batches_per_second": 1.0253790908422422,
+    "producer_headroom": 7.9797689935916845,
     "sustainable": true,
-    "buffer_seconds": 60.34419965576875,
+    "buffer_seconds": 60.46544205331264,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326244.05507966207,
-    "upload_bytes_per_second": 4822326.1971726585,
+    "generation_bytes_per_second": 340933.01408689393,
+    "upload_bytes_per_second": 4842063.584376955,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.00437964718243,
+    "estimated_full_shards_per_hour": 73.15628830866922,
     "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": 1.1908320559996355,
-      "bytes": 10104452,
-      "pages": 27,
-      "shards": 3
-    },
     {
       "seconds": 1.954254078998929,
       "bytes": 8371700,
@@ -562,6 +556,12 @@
       "bytes": 693988,
       "pages": 20,
       "shards": 1
+    },
+    {
+      "seconds": 1.3442884890027926,
+      "bytes": 12581350,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -396,3 +396,4 @@
 {"step": 194600, "loss": 0.6777424439787865, "lr": 0.0003, "grad_norm": 0.17632340123455828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28481509759, "tokens_per_second": 150106.601301932, "data_wait_s": 8.789000276010484e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 194700, "loss": 0.5743269063532352, "lr": 0.0003, "grad_norm": 0.14870336587591793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28496125885, "tokens_per_second": 131303.34703119096, "data_wait_s": 9.789997420739383e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 194800, "loss": 0.4885460287332535, "lr": 0.0003, "grad_norm": 0.15674144452471456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28510869924, "tokens_per_second": 150776.8410855801, "data_wait_s": 8.791997970547527e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 194900, "loss": 0.6080953255295753, "lr": 0.0003, "grad_norm": 0.16244985294831388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28525593721, "tokens_per_second": 150274.05671662698, "data_wait_s": 1.1323994840495288e-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 sha256:4a6d95ecd68dec70e7f1909d139cc3c090aa5dbdcc1529738578645d931207f6
-size 848610
+oid sha256:67d9448a6b40dee7744c3eb3683c1d58b161925c52b758089e8ac78d99da4d99
+size 850728