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": 76164.5364840855,
-      "required_pages_per_second_upper_estimate": 0.006601748985187645
+      "required_token_positions_per_second": 76057.42028184011,
+      "required_pages_per_second_upper_estimate": 0.006592464424260547
     },
     {
       "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": 9520.567060510688,
-      "required_pages_per_second_upper_estimate": 0.0037978965456002423
+      "required_token_positions_per_second": 9507.177535230014,
+      "required_pages_per_second_upper_estimate": 0.0037925552637745386
     },
     {
       "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": 19041.134121021376,
-      "required_pages_per_second_upper_estimate": 0.004404831266383279
+      "required_token_positions_per_second": 19014.355070460027,
+      "required_pages_per_second_upper_estimate": 0.004398636404331104
     },
     {
       "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": 28561.701181532066,
-      "required_pages_per_second_upper_estimate": 0.0017392577502903502
+      "required_token_positions_per_second": 28521.53260569004,
+      "required_pages_per_second_upper_estimate": 0.001736811694769801
     },
     {
       "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": 28561.701181532066,
-      "required_pages_per_second_upper_estimate": 0.0021987588208035205
+      "required_token_positions_per_second": 28521.53260569004,
+      "required_pages_per_second_upper_estimate": 0.0021956665326414667
     },
     {
       "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": 28561.701181532066,
-      "required_pages_per_second_upper_estimate": 0.35464073880988944
+      "required_token_positions_per_second": 28521.53260569004,
+      "required_pages_per_second_upper_estimate": 0.35414197953350685
     }
   ],
   "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": 0.0543465780937602,
-    "producer_p95_seconds": 0.15199401500012755,
-    "gpu_update_seconds": 0.9680095672269388,
-    "producer_batches_per_second": 18.40042253027914,
-    "gpu_batches_per_second": 1.033047641114441,
-    "producer_headroom": 17.811785050328325,
+    "producer_batch_seconds": 0.050597157453115216,
+    "producer_p95_seconds": 0.14450748000035674,
+    "gpu_update_seconds": 0.9693728728478016,
+    "producer_batches_per_second": 19.76395612592721,
+    "gpu_batches_per_second": 1.0315947846386733,
+    "producer_headroom": 19.158642928627966,
     "sustainable": true,
-    "buffer_seconds": 60.0165931680702,
+    "buffer_seconds": 60.1011181165637,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -18,3 +18,4 @@
 {"step": 156800, "loss": 0.6177155636250973, "lr": 0.0003, "grad_norm": 0.1492917735248138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22948289405, "tokens_per_second": 151428.1687673985, "data_wait_s": 1.3968000075692544e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156900, "loss": 0.599456112831831, "lr": 0.0003, "grad_norm": 0.14735778085006324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22963019955, "tokens_per_second": 151713.41950798957, "data_wait_s": 1.4616000044043176e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157000, "loss": 0.6236628387123346, "lr": 0.0003, "grad_norm": 0.15793851099557044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22977662403, "tokens_per_second": 150220.5779283792, "data_wait_s": 1.3168999885238009e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 157100, "loss": 0.603167487308383, "lr": 0.0003, "grad_norm": 0.13626083682581658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22992238879, "tokens_per_second": 150001.94093199095, "data_wait_s": 1.3675999980478082e-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:f99b20ffd46a6d0ff7bbc3b663c35676a51afbe68a271ba44181c184b25f033d
-size 42525
+oid sha256:f11f42f314583ca30ff1250cba71caed6cc54ecec896ffd9b2b2a8621ffc902e
+size 44643