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": 77306.57796973751,
-      "required_pages_per_second_upper_estimate": 0.0053482484399205174
+      "required_token_positions_per_second": 77244.46108045512,
+      "required_pages_per_second_upper_estimate": 0.0053439510493889205
     },
     {
       "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": 9663.322246217189,
-      "required_pages_per_second_upper_estimate": 0.004045831231375976
+      "required_token_positions_per_second": 9655.55763505689,
+      "required_pages_per_second_upper_estimate": 0.004042580350826678
     },
     {
       "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": 19326.644492434378,
-      "required_pages_per_second_upper_estimate": 0.004025230369884462
+      "required_token_positions_per_second": 19311.11527011378,
+      "required_pages_per_second_upper_estimate": 0.004021996042408214
     },
     {
       "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": 28989.966738651565,
-      "required_pages_per_second_upper_estimate": 0.0017379042408659847
+      "required_token_positions_per_second": 28966.672905170668,
+      "required_pages_per_second_upper_estimate": 0.0017365078111164292
     },
     {
       "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": 28989.966738651565,
-      "required_pages_per_second_upper_estimate": 0.002253842400542506
+      "required_token_positions_per_second": 28966.672905170668,
+      "required_pages_per_second_upper_estimate": 0.0022520314074481115
     },
     {
       "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": 28989.966738651565,
-      "required_pages_per_second_upper_estimate": 0.3718521663222838
+      "required_token_positions_per_second": 28966.672905170668,
+      "required_pages_per_second_upper_estimate": 0.3715533780373606
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4319589062500029,
-    "producer_p95_seconds": 0.2138562140007707,
-    "gpu_update_seconds": 0.9537092694603766,
-    "producer_batches_per_second": 2.315035031182421,
-    "gpu_batches_per_second": 1.0485375701190525,
-    "producer_headroom": 2.207870368364167,
+    "producer_batch_seconds": 0.1067877218827391,
+    "producer_p95_seconds": 0.21957345999908284,
+    "gpu_update_seconds": 0.9544762041022916,
+    "producer_batches_per_second": 9.364372442536743,
+    "gpu_batches_per_second": 1.0476950558872493,
+    "producer_headroom": 8.938070662752576,
     "sustainable": true,
-    "buffer_seconds": 60.08368397600373,
+    "buffer_seconds": 60.13200085844437,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323973.66989901883,
-    "upload_bytes_per_second": 5853710.994822643,
+    "generation_bytes_per_second": 324284.11406911624,
+    "upload_bytes_per_second": 5918934.498085998,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.51720783927844,
+    "estimated_full_shards_per_hour": 69.58382193141094,
     "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": 1.0810250989998167,
-      "bytes": 5909598,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9096889889988233,
       "bytes": 1202765,
@@ -562,6 +556,12 @@
       "bytes": 6087872,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.244901927000683,
+      "bytes": 11604745,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -133,3 +133,4 @@
 {"step": 498300, "loss": 0.5782458782196045, "lr": 0.0003, "grad_norm": 0.1476533573606139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72934606497, "tokens_per_second": 152606.39386160494, "data_wait_s": 9.33699993765913e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 498400, "loss": 0.5458378195762634, "lr": 0.0003, "grad_norm": 0.16699920113869673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72949286780, "tokens_per_second": 141572.82226803506, "data_wait_s": 1.2077000064891763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 498500, "loss": 0.5711762309074402, "lr": 0.0003, "grad_norm": 0.1529487524209411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72963879918, "tokens_per_second": 151916.1781062014, "data_wait_s": 1.2923999747727066e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 498600, "loss": 0.6222963929176331, "lr": 0.0003, "grad_norm": 0.1689966483910953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72978403224, "tokens_per_second": 151589.2177978077, "data_wait_s": 9.921999662765302e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ea713820685ef96a1f3a60ac36c693e92100f6638d3d2cf9d20bbc94912032d
-size 287661
+oid sha256:863887f560934a6a4600933056838e7bd8b912ea16014c4b96b38410d49d6aec
+size 289779