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": 77480.45280172658,
-      "required_pages_per_second_upper_estimate": 0.005360277504242715
+      "required_token_positions_per_second": 77624.80348000207,
+      "required_pages_per_second_upper_estimate": 0.005370264019105533
     },
     {
       "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": 9685.056600215823,
-      "required_pages_per_second_upper_estimate": 0.004054930951530282
+      "required_token_positions_per_second": 9703.100435000259,
+      "required_pages_per_second_upper_estimate": 0.0040624855283564075
     },
     {
       "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": 19370.113200431646,
-      "required_pages_per_second_upper_estimate": 0.004034283755413375
+      "required_token_positions_per_second": 19406.200870000517,
+      "required_pages_per_second_upper_estimate": 0.004041799865288725
     },
     {
       "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": 29055.16980064747,
-      "required_pages_per_second_upper_estimate": 0.001741813064873328
+      "required_token_positions_per_second": 29109.301305000776,
+      "required_pages_per_second_upper_estimate": 0.0017450581659052876
     },
     {
       "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": 29055.16980064747,
-      "required_pages_per_second_upper_estimate": 0.002258911651814728
+      "required_token_positions_per_second": 29109.301305000776,
+      "required_pages_per_second_upper_estimate": 0.002263120138178873
     },
     {
       "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": 29055.16980064747,
-      "required_pages_per_second_upper_estimate": 0.3726885211919738
+      "required_token_positions_per_second": 29109.301305000776,
+      "required_pages_per_second_upper_estimate": 0.37338286200793697
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04412602000781263,
-    "producer_p95_seconds": 0.1315502459997333,
-    "gpu_update_seconds": 0.9515690388215831,
-    "producer_batches_per_second": 22.662365647818394,
-    "gpu_batches_per_second": 1.0508958984609182,
-    "producer_headroom": 21.56480549691781,
+    "producer_batch_seconds": 0.057187040265617384,
+    "producer_p95_seconds": 0.15147459999980128,
+    "gpu_update_seconds": 0.9497995060173521,
+    "producer_batches_per_second": 17.486479372866423,
+    "gpu_batches_per_second": 1.0528537798394377,
+    "producer_headroom": 16.608649470331144,
     "sustainable": true,
-    "buffer_seconds": 60.900418484581316,
+    "buffer_seconds": 60.78716838511053,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -25,3 +25,4 @@
 {"step": 647500, "loss": 0.5511292815208435, "lr": 0.0001, "grad_norm": 0.17853451125416817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94777372465, "tokens_per_second": 154982.65392728505, "data_wait_s": 1.3123999906383688e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 647600, "loss": 0.57598477602005, "lr": 0.0001, "grad_norm": 0.17728878264916242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94792018838, "tokens_per_second": 153620.90615440658, "data_wait_s": 1.6302999938488938e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 647700, "loss": 0.5617396831512451, "lr": 0.0001, "grad_norm": 0.18557115651766962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94806559444, "tokens_per_second": 152473.34164117114, "data_wait_s": 1.5891000202827854e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 647800, "loss": 0.5407651662826538, "lr": 0.0001, "grad_norm": 0.17395694584160637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94821128922, "tokens_per_second": 152884.949190817, "data_wait_s": 1.7751000086718705e-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:d9aa82725fa7c92f5354fa6b0d84541ca34f12a7cb19af32c9aa2332c93a0b4c
-size 57351
+oid sha256:a5db7a552b35b3916708d9836fdf38944c0c12f3ecd33acf1de7978a55bb82bd
+size 59469