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": 76262.40147915448,
-      "required_pages_per_second_upper_estimate": 0.0066102316749236615
+      "required_token_positions_per_second": 76158.92228977644,
+      "required_pages_per_second_upper_estimate": 0.006601262361053981
     },
     {
       "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": 9532.80018489431,
-      "required_pages_per_second_upper_estimate": 0.003802776521818378
+      "required_token_positions_per_second": 9519.865286222055,
+      "required_pages_per_second_upper_estimate": 0.0037976165973440464
     },
     {
       "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": 19065.60036978862,
-      "required_pages_per_second_upper_estimate": 0.004410491102444309
+      "required_token_positions_per_second": 19039.73057244411,
+      "required_pages_per_second_upper_estimate": 0.00440450658012154
     },
     {
       "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": 28598.400554682932,
-      "required_pages_per_second_upper_estimate": 0.0017414925495685078
+      "required_token_positions_per_second": 28559.59585866616,
+      "required_pages_per_second_upper_estimate": 0.001739129547173592
     },
     {
       "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": 28598.400554682932,
-      "required_pages_per_second_upper_estimate": 0.002201584040139041
+      "required_token_positions_per_second": 28559.59585866616,
+      "required_pages_per_second_upper_estimate": 0.0021985967472213964
     },
     {
       "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": 28598.400554682932,
-      "required_pages_per_second_upper_estimate": 0.35509642219952237
+      "required_token_positions_per_second": 28559.59585866616,
+      "required_pages_per_second_upper_estimate": 0.35461459774595727
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04658439075000054,
-    "producer_p95_seconds": 0.12595866000003753,
-    "gpu_update_seconds": 0.9667673528501822,
-    "producer_batches_per_second": 21.466417911669016,
-    "gpu_batches_per_second": 1.0343750200623167,
-    "producer_headroom": 20.753032019639992,
+    "producer_batch_seconds": 0.04978584160157595,
+    "producer_p95_seconds": 0.1263073790000817,
+    "gpu_update_seconds": 0.9680809258234112,
+    "producer_batches_per_second": 20.08603184822621,
+    "gpu_batches_per_second": 1.0329714937306917,
+    "producer_headroom": 19.444904307749354,
     "sustainable": true,
-    "buffer_seconds": 60.90634322956148,
+    "buffer_seconds": 60.021017401051495,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 55,
+    "upload_pages_per_commit": 57,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -3,3 +3,4 @@
 {"step": 115300, "loss": 0.617220900952816, "lr": 0.0003, "grad_norm": 0.12405028442484603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16872235692, "tokens_per_second": 149967.25268749727, "data_wait_s": 1.4869999972688674e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 115400, "loss": 0.610551755875349, "lr": 0.0003, "grad_norm": 0.15634487661387214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16886971989, "tokens_per_second": 152076.35164390146, "data_wait_s": 1.957099993887823e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 115500, "loss": 0.5834813602268696, "lr": 0.0003, "grad_norm": 0.15759322127084688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16901693141, "tokens_per_second": 151467.96586301902, "data_wait_s": 1.4530000044032931e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 115600, "loss": 0.7220963202416897, "lr": 0.0003, "grad_norm": 0.16984515292624902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16916387672, "tokens_per_second": 151576.28877827647, "data_wait_s": 1.5824000001884997e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4bc66c6b379270cdf878f3298835135441158c38e4acfacdaaf69c017fb13500
-size 10755
+oid sha256:0de1f9f8c917bf904f949045375d85e5f8cbffb7f7a1089ffcf1473bee00a56c
+size 12873