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": 77377.07109584377,
-      "required_pages_per_second_upper_estimate": 0.005353125317951136
+      "required_token_positions_per_second": 76690.72546649,
+      "required_pages_per_second_upper_estimate": 0.005305642334771172
     },
     {
       "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": 9672.133886980471,
-      "required_pages_per_second_upper_estimate": 0.004049520481355579
+      "required_token_positions_per_second": 9586.34068331125,
+      "required_pages_per_second_upper_estimate": 0.004013600658545094
     },
     {
       "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": 19344.267773960943,
-      "required_pages_per_second_upper_estimate": 0.0040289008346692575
+      "required_token_positions_per_second": 19172.6813666225,
+      "required_pages_per_second_upper_estimate": 0.003993163911058517
     },
     {
       "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": 29016.40166094141,
-      "required_pages_per_second_upper_estimate": 0.0017394889740934719
+      "required_token_positions_per_second": 28759.02204993375,
+      "required_pages_per_second_upper_estimate": 0.0017240594594094734
     },
     {
       "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": 29016.40166094141,
-      "required_pages_per_second_upper_estimate": 0.0022558975994756065
+      "required_token_positions_per_second": 28759.02204993375,
+      "required_pages_per_second_upper_estimate": 0.0022358874668130264
     },
     {
       "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": 29016.40166094141,
-      "required_pages_per_second_upper_estimate": 0.37219124512180973
+      "required_token_positions_per_second": 28759.02204993375,
+      "required_pages_per_second_upper_estimate": 0.3688898558244988
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10680449580465279,
-    "producer_p95_seconds": 0.27847352100070566,
-    "gpu_update_seconds": 0.9528404080929374,
-    "producer_batches_per_second": 9.362901743658963,
-    "gpu_batches_per_second": 1.0494936943338184,
-    "producer_headroom": 8.92135111836208,
+    "producer_batch_seconds": 0.16217335250780707,
+    "producer_p95_seconds": 0.2698826400010148,
+    "gpu_update_seconds": 0.9613678779478418,
+    "producer_batches_per_second": 6.166241152052769,
+    "gpu_batches_per_second": 1.0401845359495714,
+    "producer_headroom": 5.9280261712636255,
     "sustainable": true,
-    "buffer_seconds": 60.02894570985505,
+    "buffer_seconds": 60.566176310714035,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 538184.8396180948,
-    "upload_bytes_per_second": 5303406.713660196,
+    "generation_bytes_per_second": 534355.4578108246,
+    "upload_bytes_per_second": 5313811.862840319,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.48193828017362,
+    "estimated_full_shards_per_hour": 114.66024208777957,
     "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": 1.140193093000562,
-      "bytes": 7462921,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.6433775559999049,
       "bytes": 5520234,
@@ -562,6 +556,12 @@
       "bytes": 2648981,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4223995709980954,
+      "bytes": 9800135,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 669200, "loss": 0.5672634840011597, "lr": 0.0001, "grad_norm": 0.20155257527473036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97953963073, "tokens_per_second": 152766.596153541, "data_wait_s": 1.352199979010038e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 669300, "loss": 0.6395324468612671, "lr": 0.0001, "grad_norm": 0.21508849237345326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97968717292, "tokens_per_second": 153556.8786177173, "data_wait_s": 1.1770000128308311e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 669400, "loss": 0.5363394021987915, "lr": 0.0001, "grad_norm": 0.2073551345685677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97983400795, "tokens_per_second": 151838.5984416247, "data_wait_s": 1.1027001164620742e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 669500, "loss": 0.5883676409721375, "lr": 0.0001, "grad_norm": 0.18916531821171728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97997909063, "tokens_per_second": 151322.09718859673, "data_wait_s": 1.551699824631214e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f491f8fa8091fb594b47f6ce4893784996dca640386f0e9f9f5349bfa965b8ff
-size 520089
+oid sha256:474e65e49802b8483d92e0dc8e58d712a658402f9691bae1446cab006985f387
+size 522207