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": 74968.80783643853,
-      "required_pages_per_second_upper_estimate": 0.005186516077724851
+      "required_token_positions_per_second": 75637.98173623772,
+      "required_pages_per_second_upper_estimate": 0.005232811080810327
     },
     {
       "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": 9371.100979554816,
-      "required_pages_per_second_upper_estimate": 0.003923484289298401
+      "required_token_positions_per_second": 9454.747717029715,
+      "required_pages_per_second_upper_estimate": 0.003958505431536634
     },
     {
       "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": 18742.201959109632,
-      "required_pages_per_second_upper_estimate": 0.0039035064029789863
+      "required_token_positions_per_second": 18909.49543405943,
+      "required_pages_per_second_upper_estimate": 0.0039383492219854125
     },
     {
       "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": 28113.30293866445,
-      "required_pages_per_second_upper_estimate": 0.001685349584644873
+      "required_token_positions_per_second": 28364.243151089144,
+      "required_pages_per_second_upper_estimate": 0.001700393067216215
     },
     {
       "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": 28113.30293866445,
-      "required_pages_per_second_upper_estimate": 0.0021856856461300454
+      "required_token_positions_per_second": 28364.243151089144,
+      "required_pages_per_second_upper_estimate": 0.002205195143876721
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28113.30293866445,
-      "required_pages_per_second_upper_estimate": 0.36060726438430046
+      "required_token_positions_per_second": 28364.243151089144,
+      "required_pages_per_second_upper_estimate": 0.3638260559906767
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.30113083285908715,
-    "producer_p95_seconds": 0.32759419799549505,
-    "gpu_update_seconds": 0.9834490120325025,
-    "producer_batches_per_second": 3.320815708260421,
-    "gpu_batches_per_second": 1.0168295333718333,
-    "producer_headroom": 3.265852927430726,
+    "producer_batch_seconds": 0.10701158946858413,
+    "producer_p95_seconds": 0.3155742789967917,
+    "gpu_update_seconds": 0.9747483778335315,
+    "producer_batches_per_second": 9.344782233083029,
+    "gpu_batches_per_second": 1.0259057852679812,
+    "producer_headroom": 9.10881132290529,
     "sustainable": true,
-    "buffer_seconds": 60.973838746015154,
+    "buffer_seconds": 60.434399425678954,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 499986.15021001134,
-    "upload_bytes_per_second": 5070819.72430591,
+    "generation_bytes_per_second": 504425.6926375263,
+    "upload_bytes_per_second": 5174904.206804973,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.28538875317817,
+    "estimated_full_shards_per_hour": 108.23801121086447,
     "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": 1.3267378080017806,
-      "bytes": 1199966,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 1.69540837599925,
       "bytes": 9973657,
@@ -562,6 +556,12 @@
       "bytes": 1015286,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.624823366997589,
+      "bytes": 11605166,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -374,3 +374,4 @@
 {"step": 682400, "loss": 0.5416658520698547, "lr": 0.0001, "grad_norm": 0.19516915073326702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99885489902, "tokens_per_second": 152872.42853994438, "data_wait_s": 9.359006071463227e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 682500, "loss": 0.5104610323905945, "lr": 0.0001, "grad_norm": 0.18457843788907458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99900283272, "tokens_per_second": 153702.87580703024, "data_wait_s": 1.3820004824083298e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 682600, "loss": 0.5499789118766785, "lr": 0.0001, "grad_norm": 0.19782871639603708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99914993963, "tokens_per_second": 153097.4538619784, "data_wait_s": 1.4753997675143182e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 682700, "loss": 0.510818362236023, "lr": 0.0001, "grad_norm": 0.20056314603891742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99929748758, "tokens_per_second": 153672.10828742673, "data_wait_s": 1.0219999239780009e-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 sha256:3dcdf3c6a44ef4ee7fcb2bdcabdd3e0ecd564fc3ac31bcea8fa4f65d1cddc30f
-size 802014
+oid sha256:70bfbc9278e58d29da13e417282fc503b65990c389935df4c9cc9e3c9d826f12
+size 804132