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": 75792.49342190342,
-      "required_pages_per_second_upper_estimate": 0.006569501235492222
+      "required_token_positions_per_second": 75785.50961242938,
+      "required_pages_per_second_upper_estimate": 0.006568895896587316
     },
     {
       "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": 9474.061677737927,
-      "required_pages_per_second_upper_estimate": 0.0037793448530947533
+      "required_token_positions_per_second": 9473.188701553672,
+      "required_pages_per_second_upper_estimate": 0.003778996609842697
     },
     {
       "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": 18948.123355475855,
-      "required_pages_per_second_upper_estimate": 0.004383314862707888
+      "required_token_positions_per_second": 18946.377403107344,
+      "required_pages_per_second_upper_estimate": 0.004382910967355145
     },
     {
       "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": 28422.18503321378,
-      "required_pages_per_second_upper_estimate": 0.001730761948842426
+      "required_token_positions_per_second": 28419.56610466102,
+      "required_pages_per_second_upper_estimate": 0.001730602469834019
     },
     {
       "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": 28422.18503321378,
-      "required_pages_per_second_upper_estimate": 0.002188018481500562
+      "required_token_positions_per_second": 28419.56610466102,
+      "required_pages_per_second_upper_estimate": 0.002187816869130912
     },
     {
       "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": 28422.18503321378,
-      "required_pages_per_second_upper_estimate": 0.35290841517828797
+      "required_token_positions_per_second": 28419.56610466102,
+      "required_pages_per_second_upper_estimate": 0.35287589685065274
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.08176327426554053,
-    "producer_p95_seconds": 0.21036276599988923,
-    "gpu_update_seconds": 0.9727612415333629,
-    "producer_batches_per_second": 12.230429969723634,
-    "gpu_batches_per_second": 1.0280014841295493,
-    "producer_headroom": 11.897288241835211,
+    "producer_batch_seconds": 0.06633539197656546,
+    "producer_p95_seconds": 0.25371016299868643,
+    "gpu_update_seconds": 0.9728508837249814,
+    "producer_batches_per_second": 15.074909037294503,
+    "gpu_batches_per_second": 1.0279067601512233,
+    "producer_headroom": 14.665638579005668,
     "sustainable": true,
-    "buffer_seconds": 60.3111969750685,
+    "buffer_seconds": 60.31675479094885,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278162.4472644019,
-    "upload_bytes_per_second": 5795361.617427259,
+    "generation_bytes_per_second": 276546.64650794666,
+    "upload_bytes_per_second": 5847206.49680464,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.687185892572806,
+    "estimated_full_shards_per_hour": 59.34047266415405,
     "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": 0.8255870570001207,
-      "bytes": 1120534,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.3186273349992916,
       "bytes": 13209471,
@@ -562,6 +556,12 @@
       "bytes": 879863,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.04706289600108,
+      "bytes": 6208354,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -144,3 +144,4 @@
 {"step": 369400, "loss": 0.589466156437993, "lr": 0.0003, "grad_norm": 0.16202583969873532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54069685623, "tokens_per_second": 149994.51488694124, "data_wait_s": 1.4674000340164639e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 369500, "loss": 0.6001040088012815, "lr": 0.0003, "grad_norm": 0.15605304812775767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54084430133, "tokens_per_second": 150554.46778938608, "data_wait_s": 9.348999810754322e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 369600, "loss": 0.6240045689046383, "lr": 0.0003, "grad_norm": 0.15953225910088997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54099157674, "tokens_per_second": 150407.05664663756, "data_wait_s": 9.65000072028488e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 369700, "loss": 0.613152289763093, "lr": 0.0003, "grad_norm": 0.1619095578312672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54113773186, "tokens_per_second": 148900.5519914343, "data_wait_s": 1.0379999366705306e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ecc773a8a71a55804f71956ec0f1ab56b87d8ed86c58b57c84843c483f87b43b
-size 310959
+oid sha256:262142a3f0ebb94f35e0976e39363ad5ec54d0e7097a9c69f4993aea731e59b4
+size 313077