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": 76389.5922770699,
-      "required_pages_per_second_upper_estimate": 0.0052848092420018245
+      "required_token_positions_per_second": 76356.00858976411,
+      "required_pages_per_second_upper_estimate": 0.00528248584458912
     },
     {
       "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": 9548.699034633737,
-      "required_pages_per_second_upper_estimate": 0.003997840886290828
+      "required_token_positions_per_second": 9544.501073720514,
+      "required_pages_per_second_upper_estimate": 0.003996083287719854
     },
     {
       "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": 19097.398069267474,
-      "required_pages_per_second_upper_estimate": 0.003977484385573526
+      "required_token_positions_per_second": 19089.002147441028,
+      "required_pages_per_second_upper_estimate": 0.003975735736472426
     },
     {
       "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": 28646.097103901207,
-      "required_pages_per_second_upper_estimate": 0.0017172897813212311
+      "required_token_positions_per_second": 28633.503221161543,
+      "required_pages_per_second_upper_estimate": 0.0017165347972807339
     },
     {
       "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": 28646.097103901207,
-      "required_pages_per_second_upper_estimate": 0.0022271080489633405
+      "required_token_positions_per_second": 28633.503221161543,
+      "required_pages_per_second_upper_estimate": 0.0022261289299749677
     },
     {
       "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": 28646.097103901207,
-      "required_pages_per_second_upper_estimate": 0.36744137586615366
+      "required_token_positions_per_second": 28633.503221161543,
+      "required_pages_per_second_upper_estimate": 0.367279835060627
     }
   ],
   "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": 0.05945815298437651,
-    "producer_p95_seconds": 0.19198795900047116,
-    "gpu_update_seconds": 0.9651576582917719,
-    "producer_batches_per_second": 16.818551364398495,
-    "gpu_batches_per_second": 1.0361001556677232,
-    "producer_headroom": 16.232553650722735,
+    "producer_batch_seconds": 0.07203993300774414,
+    "producer_p95_seconds": 0.18900574699910067,
+    "gpu_update_seconds": 0.9655821638885874,
+    "producer_batches_per_second": 13.881190032374157,
+    "gpu_batches_per_second": 1.035644647756132,
+    "producer_headroom": 13.40342950880853,
     "sustainable": true,
-    "buffer_seconds": 60.80493247238163,
+    "buffer_seconds": 60.83167632498101,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 89,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 452285.9608548129,
+    "generation_bytes_per_second": 485026.75635320565,
     "upload_bytes_per_second": 5029592.1244579265,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.05003852113046,
+    "estimated_full_shards_per_hour": 104.07545106837394,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -92,3 +92,4 @@
 {"step": 406300, "loss": 0.6408392786979675, "lr": 0.0003, "grad_norm": 0.18752841436474124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59471756710, "tokens_per_second": 152487.0561394124, "data_wait_s": 9.763001799001358e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 406400, "loss": 0.5909572243690491, "lr": 0.0003, "grad_norm": 0.1529850649040701, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59486498541, "tokens_per_second": 153533.83831338357, "data_wait_s": 1.0314999599358998e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 406500, "loss": 0.5848004221916199, "lr": 0.0003, "grad_norm": 0.16868212642409153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59501288633, "tokens_per_second": 154018.16408604814, "data_wait_s": 1.0071000360767357e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 406600, "loss": 0.6084566116333008, "lr": 0.0003, "grad_norm": 0.16289168414628516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59515798258, "tokens_per_second": 151487.60826619124, "data_wait_s": 9.319001037511043e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0078ebf2f173170a188a0af14ab897b07141db68f79e5c0a57843b6033edc8c8
-size 200823
+oid sha256:e9de710a9587d8c2fac185892900c656c8b8bcc2ad905fd75d2aa3bd17bcae24
+size 202941