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": 75639.18243854106,
-      "required_pages_per_second_upper_estimate": 0.006556212628017507
+      "required_token_positions_per_second": 75740.92127633149,
+      "required_pages_per_second_upper_estimate": 0.006565031092622455
     },
     {
       "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": 9454.897804817632,
-      "required_pages_per_second_upper_estimate": 0.003771700097661414
+      "required_token_positions_per_second": 9467.615159541436,
+      "required_pages_per_second_upper_estimate": 0.0037767732406021365
     },
     {
       "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": 18909.795609635265,
-      "required_pages_per_second_upper_estimate": 0.004374448413252956
+      "required_token_positions_per_second": 18935.23031908287,
+      "required_pages_per_second_upper_estimate": 0.004380332285648062
     },
     {
       "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": 28364.693414452897,
-      "required_pages_per_second_upper_estimate": 0.0017272610108951003
+      "required_token_positions_per_second": 28402.84547862431,
+      "required_pages_per_second_upper_estimate": 0.0017295842714347284
     },
     {
       "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": 28364.693414452897,
-      "required_pages_per_second_upper_estimate": 0.0021835926175413625
+      "required_token_positions_per_second": 28402.84547862431,
+      "required_pages_per_second_upper_estimate": 0.002186529669052951
     },
     {
       "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": 28364.693414452897,
-      "required_pages_per_second_upper_estimate": 0.3521945616853483
+      "required_token_positions_per_second": 28402.84547862431,
+      "required_pages_per_second_upper_estimate": 0.35266828263561234
     }
   ],
   "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": 0.10634636189061553,
-    "producer_p95_seconds": 0.36112446799961617,
-    "gpu_update_seconds": 0.9747329046014749,
-    "producer_batches_per_second": 9.403236577369409,
-    "gpu_batches_per_second": 1.0259220708352466,
-    "producer_headroom": 9.165644101714115,
+    "producer_batch_seconds": 0.11075003331254152,
+    "producer_p95_seconds": 0.3574174440000206,
+    "gpu_update_seconds": 0.9734235966184305,
+    "producer_batches_per_second": 9.029342656520523,
+    "gpu_batches_per_second": 1.0273019921377426,
+    "producer_headroom": 8.78937520381042,
     "sustainable": true,
-    "buffer_seconds": 60.433440085291444,
+    "buffer_seconds": 60.35226299034269,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330840.66579342773,
-    "upload_bytes_per_second": 2740209.0460170023,
+    "generation_bytes_per_second": 331521.35851959203,
+    "upload_bytes_per_second": 2750203.136359872,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.99070530273556,
+    "estimated_full_shards_per_hour": 71.13676611605473,
     "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": 1.675639938999666,
-      "bytes": 7647558,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.968150565000542,
       "bytes": 1141925,
@@ -562,6 +556,12 @@
       "bytes": 5465773,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.024480873002176,
+      "bytes": 7386490,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -268,3 +268,4 @@
 {"step": 221800, "loss": 0.5743122920393944, "lr": 0.0003, "grad_norm": 0.15398138982626403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32464542690, "tokens_per_second": 147316.69336070892, "data_wait_s": 1.1284999345662072e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 221900, "loss": 0.5680422894656658, "lr": 0.0003, "grad_norm": 0.14915218842361433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32479159318, "tokens_per_second": 148124.76639618704, "data_wait_s": 1.5765999705763534e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 222000, "loss": 0.6181822307407856, "lr": 0.0003, "grad_norm": 0.14457738280646545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32494021399, "tokens_per_second": 151195.13838724297, "data_wait_s": 8.519000402884558e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 222100, "loss": 0.5670797917991877, "lr": 0.0003, "grad_norm": 0.177642068664115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32508712610, "tokens_per_second": 149269.8434673818, "data_wait_s": 8.057999366428703e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:29f3c053e31ac400f278dfacab6ed0baf823a81767e1b8ec2214676e9ec44faa
-size 575940
+oid sha256:f5bc6a636916466fc56cdc6a9132475b774c04f17356d88a400c120cb09a4ecb
+size 578058