Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77665.39252487822,
-      "required_pages_per_second_upper_estimate": 0.005373072063409624
+      "required_token_positions_per_second": 77311.1278608318,
+      "required_pages_per_second_upper_estimate": 0.005348563211943621
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9708.174065609777,
-      "required_pages_per_second_upper_estimate": 0.004064609751543158
+      "required_token_positions_per_second": 9663.890982603974,
+      "required_pages_per_second_upper_estimate": 0.004046069349424556
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19416.348131219554,
-      "required_pages_per_second_upper_estimate": 0.004043913272199376
+      "required_token_positions_per_second": 19327.78196520795,
+      "required_pages_per_second_upper_estimate": 0.004025467275466023
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29124.522196829334,
-      "required_pages_per_second_upper_estimate": 0.0017459706351294535
+      "required_token_positions_per_second": 28991.672947811923,
+      "required_pages_per_second_upper_estimate": 0.0017380065254999177
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29124.522196829334,
-      "required_pages_per_second_upper_estimate": 0.0022643034955689154
+      "required_token_positions_per_second": 28991.672947811923,
+      "required_pages_per_second_upper_estimate": 0.002253975050799903
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29124.522196829334,
-      "required_pages_per_second_upper_estimate": 0.37357809926539337
+      "required_token_positions_per_second": 28991.672947811923,
+      "required_pages_per_second_upper_estimate": 0.3718740517414082
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.15859014459383047,
-    "producer_p95_seconds": 0.2407994429995597,
-    "gpu_update_seconds": 0.9493031272118148,
-    "producer_batches_per_second": 6.305562067309587,
-    "gpu_batches_per_second": 1.0534043039941166,
-    "producer_headroom": 5.985889789325187,
+    "producer_batch_seconds": 0.06760819719525557,
+    "producer_p95_seconds": 0.24046512700078893,
+    "gpu_update_seconds": 0.9536531420511443,
+    "producer_batches_per_second": 14.79110583457734,
+    "gpu_batches_per_second": 1.048599281966577,
+    "producer_headroom": 14.105584553555692,
     "sustainable": true,
-    "buffer_seconds": 60.755400141556144,
+    "buffer_seconds": 60.08014794922209,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 475897.6600359646,
-    "upload_bytes_per_second": 5491742.388941373,
+    "generation_bytes_per_second": 475506.9435707584,
+    "upload_bytes_per_second": 5473990.180463189,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.11655951317982,
+    "estimated_full_shards_per_hour": 102.03272085516036,
     "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": 0.9495372719993611,
-      "bytes": 5544847,
-      "pages": 18,
-      "shards": 2
-    },
     {
       "seconds": 1.068828384999506,
       "bytes": 2455135,
@@ -562,6 +556,12 @@
       "bytes": 12940523,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.308492041000136,
+      "bytes": 6108009,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 547900, "loss": 0.5492777824401855, "lr": 0.0003, "grad_norm": 0.16366448976048606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80195120841, "tokens_per_second": 151511.10119357446, "data_wait_s": 1.6111000149976462e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 548000, "loss": 0.5742793083190918, "lr": 0.0003, "grad_norm": 0.16188139865888318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80209825350, "tokens_per_second": 153958.7221827193, "data_wait_s": 8.487000741297379e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 548100, "loss": 0.5568685531616211, "lr": 0.0003, "grad_norm": 0.18329008408751127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80224534705, "tokens_per_second": 153233.11600300542, "data_wait_s": 1.5513000107603148e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 548200, "loss": 0.5650813579559326, "lr": 0.0003, "grad_norm": 0.1473897733160872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80239193557, "tokens_per_second": 152586.16318734127, "data_wait_s": 1.5380999684566632e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:85f2d2ab6b3f6b48f3b2c5463a0a2538c5d631094d0b3f167823bfa0d22b83db
-size 492555
+oid sha256:8cdf883bc98507b1448a133292b0f2973dac751db131784d86fec4f620da5de6
+size 494673