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": 75290.85293262791,
-      "required_pages_per_second_upper_estimate": 0.006526020309278018
+      "required_token_positions_per_second": 75508.89023026133,
+      "required_pages_per_second_upper_estimate": 0.0065449192296264036
     },
     {
       "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": 9411.356616578489,
-      "required_pages_per_second_upper_estimate": 0.003754330866674042
+      "required_token_positions_per_second": 9438.611278782666,
+      "required_pages_per_second_upper_estimate": 0.003765203158920802
     },
     {
       "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": 18822.713233156977,
-      "required_pages_per_second_upper_estimate": 0.004354303438052185
+      "required_token_positions_per_second": 18877.222557565332,
+      "required_pages_per_second_upper_estimate": 0.004366913210922713
     },
     {
       "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": 28234.069849735464,
-      "required_pages_per_second_upper_estimate": 0.0017193067211326877
+      "required_token_positions_per_second": 28315.833836348,
+      "required_pages_per_second_upper_estimate": 0.001724285718934907
     },
     {
       "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": 28234.069849735464,
-      "required_pages_per_second_upper_estimate": 0.0021735368539402447
+      "required_token_positions_per_second": 28315.833836348,
+      "required_pages_per_second_upper_estimate": 0.002179831271966879
     },
     {
       "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": 28234.069849735464,
-      "required_pages_per_second_upper_estimate": 0.35057265418050665
+      "required_token_positions_per_second": 28315.833836348,
+      "required_pages_per_second_upper_estimate": 0.3515878892477743
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.03829906017973883,
-    "producer_p95_seconds": 0.12654167199980293,
-    "gpu_update_seconds": 0.9792424594521942,
-    "producer_batches_per_second": 26.11030127911664,
-    "gpu_batches_per_second": 1.0211975495419365,
-    "producer_headroom": 25.568315641599952,
+    "producer_batch_seconds": 0.04027617505462899,
+    "producer_p95_seconds": 0.13305566999952134,
+    "gpu_update_seconds": 0.9764148271173027,
+    "producer_batches_per_second": 24.828574179242196,
+    "gpu_batches_per_second": 1.0241548696595775,
+    "producer_headroom": 24.24298796479389,
     "sustainable": true,
-    "buffer_seconds": 60.71303248603604,
+    "buffer_seconds": 60.537719281272764,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -32,3 +32,4 @@
 {"step": 88200, "loss": 0.6163171380758286, "lr": 0.0003, "grad_norm": 0.13880462622242368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12907718727, "tokens_per_second": 149249.81466154626, "data_wait_s": 1.227999928232748e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 88300, "loss": 0.6628314591944218, "lr": 0.0003, "grad_norm": 0.13338610566072232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12922422282, "tokens_per_second": 150240.00756796182, "data_wait_s": 1.2837001122534275e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 88400, "loss": 0.6176230721175671, "lr": 0.0003, "grad_norm": 0.12480244575333811, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12937137368, "tokens_per_second": 150555.39858962144, "data_wait_s": 1.164599962066859e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 88500, "loss": 0.6561363190412521, "lr": 0.0003, "grad_norm": 0.1666916385605709, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12951962743, "tokens_per_second": 151715.80179587103, "data_wait_s": 7.917999027995393e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2f1744c24f1d5d61ae62be033ab2e401d7f2bf3f346614aca144b28a1a8ec885
-size 72177
+oid sha256:660f8fd6c58a46bad8bd56fbbc3836622bdc5a682e7f1bb5c78f8f7d0beea46f
+size 74295