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": 77074.60309380818,
-      "required_pages_per_second_upper_estimate": 0.005332199879747857
+      "required_token_positions_per_second": 76632.40525528403,
+      "required_pages_per_second_upper_estimate": 0.00530160760723841
     },
     {
       "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": 9634.325386726023,
-      "required_pages_per_second_upper_estimate": 0.004033690851830307
+      "required_token_positions_per_second": 9579.050656910504,
+      "required_pages_per_second_upper_estimate": 0.004010548476724164
     },
     {
       "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": 19268.650773452046,
-      "required_pages_per_second_upper_estimate": 0.004013151807617659
+      "required_token_positions_per_second": 19158.101313821007,
+      "required_pages_per_second_upper_estimate": 0.003990127270561816
     },
     {
       "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": 28902.976160178066,
-      "required_pages_per_second_upper_estimate": 0.001732689288513421
+      "required_token_positions_per_second": 28737.15197073151,
+      "required_pages_per_second_upper_estimate": 0.0017227483815549765
     },
     {
       "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": 28902.976160178066,
-      "required_pages_per_second_upper_estimate": 0.0022470792656973087
+      "required_token_positions_per_second": 28737.15197073151,
+      "required_pages_per_second_upper_estimate": 0.0022341871643513644
     },
     {
       "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": 28902.976160178066,
-      "required_pages_per_second_upper_estimate": 0.37073634458483173
+      "required_token_positions_per_second": 28737.15197073151,
+      "required_pages_per_second_upper_estimate": 0.36860932993075396
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.056893671734371765,
-    "producer_p95_seconds": 0.19020467300106247,
-    "gpu_update_seconds": 0.9565796908517972,
-    "producer_batches_per_second": 17.576647270523402,
-    "gpu_batches_per_second": 1.0453912094971813,
-    "producer_headroom": 16.81346381224836,
+    "producer_batch_seconds": 0.07306959714837546,
+    "producer_p95_seconds": 0.20791816100063443,
+    "gpu_update_seconds": 0.9620995159213829,
+    "producer_batches_per_second": 13.685582499782985,
+    "gpu_batches_per_second": 1.0393935174599072,
+    "producer_headroom": 13.16689229814336,
     "sustainable": true,
-    "buffer_seconds": 60.26452052366322,
+    "buffer_seconds": 60.61226950304712,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 481847.4817296166,
+    "generation_bytes_per_second": 481492.81351413555,
     "upload_bytes_per_second": 5568003.932874464,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.39325274387716,
+    "estimated_full_shards_per_hour": 103.31714920108843,
     "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": [

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -99,3 +99,4 @@
 {"step": 567000, "loss": 0.5657712817192078, "lr": 0.0003, "grad_norm": 0.1702218418641484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82991975291, "tokens_per_second": 151759.05053646342, "data_wait_s": 8.511000487487763e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 567100, "loss": 0.5542783141136169, "lr": 0.0003, "grad_norm": 0.1549565833820938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83006501378, "tokens_per_second": 151108.9335208617, "data_wait_s": 1.4715000361320563e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 567200, "loss": 0.5882192254066467, "lr": 0.0003, "grad_norm": 0.17536141640391117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83020991332, "tokens_per_second": 151221.15868952294, "data_wait_s": 1.0689000191632658e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 567300, "loss": 0.587693452835083, "lr": 0.0003, "grad_norm": 0.18768873340360923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83035603739, "tokens_per_second": 152396.77775143608, "data_wait_s": 1.4213001122698188e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4f47ecbcd8b8ea5b631361f974a78bc7c600a319852b446e9b6fa87a4bae79a
-size 215649
+oid sha256:ae4b3f080fd3dce9fff0bdd7dbf63dccc304001d4b96103797d0775b0246b10e
+size 217767